A system artificial intelligence (AI) agent is trained to act on behalf of an online system. The system AI agent comprises a large language model that has been pre-trained using a set of system constraints and a set of system objectives. The system AI agent is trained adversarially using training service requests from a plurality of different user AI agents of different types to determine resolutions to the training service requests. Once trained, the system AI agent may determine resolutions to service requests of users of the online system. In some embodiments, the system agent may determine the resolutions via messaging with user AI agents that represent the users. The online system may further train the system AI agent (and in some embodiments the user AI agents) based in part on the resolutions to the service requests.
An online system that maintains a website, such as a white-labeled website, designed by an entity retrieves a set of contextual data associated with the website, in which the set of contextual data includes information describing the entity, one or more elements of the website, or a historical performance of the website. The online system generates a prompt including the set of contextual data and a request for a set of recommendations for improving a performance of the website by updating a set of elements of the website. The online system provides the prompt to a large language model to obtain an output and extracts, from the output, the set of recommendations for improving the performance of the website. The online system sends the set of recommendations to a computing system associated with the entity.
An online system trains a ceiling prediction model to determine a user's ceiling for one or more item categories. The user's ceiling for an item category is a maximum amount of an item within the item category the user is likely to include in an order. Based on previously fulfilled orders for the user, information describing a current order from the user, and contextual information about the order, the ceiling prediction model determines the user's ceiling for an item category. The online system leverages the user's ceiling for an item category to refine content about different items that is selected for presentation to a user. For example, the online system determines whether the order includes a quantity of items from an item category that equals the user's ceiling for the item category when determining which items to present to the user.
An agentic model supported by language models tuned for interaction with pickers on behalf of users of an online system. Upon receiving a message from a picker related to fulfillment of an order of a user, the online system selects a language model of the agentic model associated with a cluster of users including the user and tuned to have a persona of the user that is common to the cluster of users. The online system requests the language model to generate, based on a prompt input into the language model including the message from the picker, first data related to the user and second data related to the cluster of users, a response to the message on behalf of the user. The online system causes a user interface of the device of the picker and a user interface of a device associated with the user to display the response.
A smart cart system accounts for edge cases in user interactions by leveraging sensor data and machine-learning models of a smart cart system. For example, a smart cart system uses sensor data to detect when a user removes an item from the smart cart system and presents content to the user on a display of the smart cart system based on the removed item. The smart cart system captures images of the storage area and applies an item identification model to the images to identify the item removed from the storage area. The smart cart system identifies a set of candidate items based on location sensor data describing a location of the smart cart system when the item was removed and computes presentation scores for each of the set of candidate items based on item data for each item the removed item.
Embodiments relate to utilizing an optical character recognition extraction and a large language model (LLM) to automatically populate a shopping cart of a user of an online system with items from a physical recipe. The online system receives an image capturing the physical recipe and extracts a raw text from the received image. The online system generates a prompt for input into the LLM, the prompt including a task request for the LLM to generate a list of ingredients using the raw text. The online system inputs the prompt into the LLM to generate the list of ingredients. The online system maps the list of ingredients to a list of items available by one or more retailers associated with the online system. The online system causes a device of the user to display a user interface with the list of items.
An online system configures one or more system AI agent instances that interact with user AI agents and performs one or more tasks on behalf of the online system. Thus, responsive to detecting the presence of a user AI agent representing a particular user, the online system directs the session for the user to communicate and interact with a system AI agent.
An online system receives an image captured at a source location, in which the image depicts one or more objects. The system generates a prompt including the image and a request to identify, from the objects, a set of items available at the source location based on a database of items available at the source location, and to extract, from the image, text describing a price or a promotion associated with each identified item. The system provides the prompt to a large language model to obtain an output, in which the model is fine-tuned based on the database of items. The system extracts, from the output, an identifier and the text associated with each item, retrieves item data for each item based on the identifier associated with the item, and generates promotional content for the source location based on the item data and the price or promotion associated with each item.
G06V 10/44 - Local feature extraction by analysis of parts of the pattern, e.g. by detecting edges, contours, loops, corners, strokes or intersectionsConnectivity analysis, e.g. of connected components
G06V 10/25 - Determination of region of interest [ROI] or a volume of interest [VOI]
9.
Natural Language Processing for Extracting Specific Items from a List of Ingredients
An online system receives a list of ingredients and corresponding quantities of each ingredient. Based on an item catalog of specific items offered by a source, the online system retrieves items offered by the source matching the ingredients and selects an item for an ingredient. Because the source may not offer an item in the same quantity specified by the list of items, the online system also maps a quantity of an ingredient in the list to a quantity of the selected item in a unit in which the source offers the corresponding item. The online system may convert a quantity of an ingredient to a quantity of an item through application of one or more rules or through application of one or more trained models to the quantity of the ingredient.
An online system retrieves item data for items available at sources in multiple regions and generates candidate nodes based on the item data, in which each candidate node represents items having at least a threshold measure of similarity to each other. The system accesses and applies a machine-learning model to predict a matching score for each combination of an item and a candidate node based on item data for the item and attributes of items represented by the candidate node. The system assigns the items to candidate nodes based on the matching scores, retrieves information describing an availability of each item in each geographical region, and identifies an average availability of items assigned to each candidate node across the geographical regions. The system selects nodes to include in a region- and source-agnostic item database, in which the average availability associated with each selected node is at least a threshold.
An online system receives an image captured at a source location, in which the image depicts one or more objects. The system generates a prompt including the image and a request to identify, from the objects, a set of items available at the source location based on a database of items available at the source location, and to extract, from the image, text describing a price or a promotion associated with each identified item. The system provides the prompt to a large language model to obtain an output, in which the model is fine-tuned based on the database of items. The system extracts, from the output, an identifier and the text associated with each item, retrieves item data for each item based on the identifier associated with the item, and generates promotional content for the source location based on the item data and the price or promotion associated with each item.
Item linked recipe generation using machine learning is described. Raw data is received that describes a recipe that uses ingredients. Ingredient descriptors are extracted from the raw data for the ingredients. Parsed ingredient data is determined using the ingredient descriptors and a large language model, such that the parsed ingredient data for each ingredient includes a name, a quantity, and a unit of measure. The name of each ingredient is mapped to a corresponding ingredient identifier that is part of an ingredient database. And each ingredient identifier in the ingredient database is associated with a corresponding item that is available for sale at one or more sources. A linked recipe is generated that includes for each ingredient: an ingredient identifier, a quantity of the ingredient, and a unit of measure of the quantity. A recommendation for the linked recipe is provided to a user client device.
An online system performs flyer quality assurance monitoring to identify and remedy errors in flyers. The online system generates a prompt for a large language machine-learning model (LLM) to verify the flyer's accuracy. The prompt includes a portion of the flyer and a query to identify errors in that portion. The online system provides the prompt to a model serving system for execution by the LLM. The online system receives, from the model serving system, a response indicating error(s) identified in the portion of the flyer. Responsive to receiving identifying the errors, the online system performs remedial measure(s) to correct the identified error(s). Remedial measures may include correcting associations to items in an item catalog, modifying textual information or image data in the flyer, etc. The online system transmits the corrected flyer to client device(s) for presentation to user(s) of the online system.
G06F 3/0484 - Interaction techniques based on graphical user interfaces [GUI] for the control of specific functions or operations, e.g. selecting or manipulating an object, an image or a displayed text element, setting a parameter value or selecting a range
G06T 11/60 - Editing figures and textCombining figures or text
G06V 10/26 - Segmentation of patterns in the image fieldCutting or merging of image elements to establish the pattern region, e.g. clustering-based techniquesDetection of occlusion
An online system generates images for collections of items using an image generation model. To ensure a generated image accurately reflects a collection of items, the online system determines a type of the collection and selects a template including evaluation questions associated with the determined type. Evaluation questions are curated to determine accuracy of the content of a generated image for the collection. By applying a visual learning model to the questions in the selected template and the generated image, the online system identifies discrepancies between the image and the collection of items from the output of the vision language model. Subsequently, the online system prompts the image generation model to create an updated image for the collection that does not include the identified discrepancies. The online system may repeat the discrepancy identification and image modification until no discrepancies are found in the generated image.
G06T 11/60 - Editing figures and textCombining figures or text
G06V 10/44 - Local feature extraction by analysis of parts of the pattern, e.g. by detecting edges, contours, loops, corners, strokes or intersectionsConnectivity analysis, e.g. of connected components
15.
GENERATION AND ASSIGNMENT OF EXPIRATION STATUS CHECKING TASKS USING A MACHINE LEARNING MODEL TO PREDICT ITEM FRESHNESS
Generation and assignment of expiration status checking tasks using an item freshness model is described. Candidate perishable items are identified to check for expiration at a source location associated with a source computing system. The candidate perishable items are applied to an item freshness model to generate scores for the plurality of candidate perishable items. Based in part on the scores, one or more of the candidate perishable items are selected as one or more perishable items for a picker to check for expiration status. Instructions are provided to a picker client device associated with the picker to check the one or more perishable items for expiration status. Expiration status data is received from the picker client device describing whether each of the one or more perishable items are expired. The expiration status data is provided to the source computing system.
Disclosed are technologies for generating training data for identification neural networks. Series of images are captured of a plurality of merchandise items from different angles and with different background assortments of other merchandise items. A labeled training dataset is generated for the plurality of merchandise items. The series of captured images is normalized, where the merchandise occupies a threshold percentage of pixels in the normalized image. The training dataset is extended by applying augmentation operations to the normalized images to generate a plurality of augmented images. Each image is stored in the training dataset as a unique training data point for the given merchandise item it depicts. Labels are generated mapping each training data point to attributes associated with the depicted merchandise item. Input neural networks are trained on the labeled training dataset to perform real-time identification of selected merchandise items placed into a self-checkout apparatus by a user.
G06V 10/44 - Local feature extraction by analysis of parts of the pattern, e.g. by detecting edges, contours, loops, corners, strokes or intersectionsConnectivity analysis, e.g. of connected components
G06V 10/772 - Determining representative reference patterns, e.g. averaging or distorting patternsGenerating dictionaries
G06V 10/82 - Arrangements for image or video recognition or understanding using pattern recognition or machine learning using neural networks
G06V 20/20 - ScenesScene-specific elements in augmented reality scenes
17.
AUTOMATICALLY ESTABLISHING SESSIONS BETWEEN USERS AND SHOPPING CARTS
An automated checkout system automatically establishes sessions between users and shopping carts by correlating action events with distances of the user’s client device to the shopping cart. The automated checkout system determines the client device’s distance from the shopping cart at timestamps when an action event occurs with respect cart. If the distances and the action events are correlated, the system establishes a session between the user and the shopping cart. Additionally, the automated checkout system attributes target actions to recipe suggestions. The automated checkout system displays a recipe suggestion to a user on a display of a shopping cart, and identifies an item added to the shopping cart. If the added item matches an item in the set of recipes, the automated checkout system applies an attribution model that determines whether to attribute a target action that relates to the item with the recipe suggestion.
An online system generates images for collections of items using an image generation model. To ensure a generated image accurately reflects a collection of items, the online system determines a type of the collection and selects a template including evaluation questions associated with the determined type. Evaluation questions are curated to determine accuracy of the content of a generated image for the collection. By applying a visual learning model to the questions in the selected template and the generated image, the online system identifies discrepancies between the image and the collection of items from the output of the vision language model. Subsequently, the online system prompts the image generation model to create an updated image for the collection that does not include the identified discrepancies. The online system may repeat the discrepancy identification and image modification until no discrepancies are found in the generated image.
G06V 10/762 - Arrangements for image or video recognition or understanding using pattern recognition or machine learning using clustering, e.g. of similar faces in social networks
A smart cart system accounts for edge cases in user interactions by leveraging sensor data and machine-learning models of a smart cart system. For example, a smart cart system uses sensor data to detect when a user removes an item from the smart cart system and presents content to the user on a display of the smart cart system based on the removed item. The smart cart system captures images of the storage area and applies an item identification model to the images to identify the item removed from the storage area. The smart cart system identifies a set of candidate items based on location sensor data describing a location of the smart cart system when the item was removed and computes presentation scores for each of the set of candidate items based on item data for each item the removed item.
B62B 3/14 - Hand carts having more than one axis carrying transport wheelsSteering devices thereforEquipment therefor characterised by provisions for nesting or stacking, e.g. shopping trolleys
A47F 9/04 - Check-out counters, e.g. for self-service stores
B62B 5/00 - Accessories or details specially adapted for hand carts
A smart cart presents candidate content objects to a user according to presentation constraints determined based on a predicted route of the smart cart. The smart cart obtains, from an item database, a plurality of candidate content objects to be presented to a user of a smart cart. The smart cart obtains a location of the smart cart in an environment. The smart cart applies a machine-learning route prediction model to the location of the smart cart to determine a future route of the smart cart. The smart cart determines, for each candidate content object, one or more presentation constraints based on the future route of the smart cart, wherein the presentation constraints constrain presentation of the candidate content object to the user to maximize a likelihood of the user engaging with the content object. The smart cart presents, via an electronic display, one or more of the candidate content objects according to the presentation constraints.
An online system uses a trained machine-learning model for efficient packing of items. Upon receiving, from a device of an agent or a device of a source via a network, a signal indicating that a set of items are ready for packing, the online system applies the machine-learning model to identify, based at least in part on input data, a packing order for one or more items of the set of items. Based on the identified packing order for the one or more items, the online system generates a packing interface signal. The online system sends the packing interface signal, wherein sending the packing interface signal causes the one or more items to be packed according to the identified packing order. This process is repeated until it is confirmed that all items from the set of items were packed.
An online system utilizes a generative machine-learning model to generate a user interface of the online system with visualization of items of specific quantities. Upon receiving an interaction with an item on the user interface, the online system identifies a quantity of the item to show in the user interface. Responsive to identifying the quantity of the item, the online system generates a prompt for the generative model, the prompt including the identified quantity of the item, information about a reference object, and a request for generating an image of the identified quantity of the item in the reference object. The online system requests the generative model to generate, by providing the prompt to the generative model, the image of the identified quantity of the item. The online system updates the user interface to display the generated image of the identified quantity of the item in the reference object.
G06F 3/0484 - Interaction techniques based on graphical user interfaces [GUI] for the control of specific functions or operations, e.g. selecting or manipulating an object, an image or a displayed text element, setting a parameter value or selecting a range
G06F 3/0482 - Interaction with lists of selectable items, e.g. menus
G06F 9/451 - Execution arrangements for user interfaces
G06F 40/40 - Processing or translation of natural language
G06V 10/764 - Arrangements for image or video recognition or understanding using pattern recognition or machine learning using classification, e.g. of video objects
23.
PERSONALIZED MACHINE-LEARNED LARGE LANGUAGE MODEL (LLM)
A computer system finetunes a machine-learned language model to generate a personalized response to a user request. The system may generate a user representation for each of a plurality of users by applying a transformer model to a sequence of tokens representing a sequence of activities of the user. The system may train an evaluation model coupled to receive a user representation and a response to a user request and generate an estimated evaluation score indicating a level of personalization of the response to the user. The system may finetune a first machine-learned language model to generate a second machine-learned language model. The finetuned machine-learned language model is configured to provide personalized responses for customer services at an online concierge system.
An online system displays items to a user in search results based on appeasement scores for the items, adjusted according to how specific the search query is. The online system receives a search query from a user of an online system. The online system computes a query specificity score, a measure of the specificity of the search query. The online system accesses candidate items from a database that potentially match the search query. For each candidate item, the online system may compute or predict an appeasement score. The online system adjusts the appeasement score based on the query specificity score such that a more specific query weights the appeasement score lower than a less specific query. The online system may then compute a ranking score based on the adjusted appeasement score and display the candidate items to the user based on their ranking scores.
A smart cart presents candidate content objects to a user according to presentation constraints determined based on a predicted route of the smart cart. The smart cart obtains, from an item database, a plurality of candidate content objects to be presented to a user of a smart cart. The smart cart obtains a location of the smart cart in an environment. The smart cart applies a machine-learning route prediction model to the location of the smart cart to determine a future route of the smart cart. The smart cart determines, for each candidate content object, one or more presentation constraints based on the future route of the smart cart, wherein the presentation constraints constrain presentation of the candidate content object to the user to maximize a likelihood of the user engaging with the content object. The smart cart presents, via an electronic display, one or more of the candidate content objects according to the presentation constraints.
B62B 5/00 - Accessories or details specially adapted for hand carts
H04W 4/02 - Services making use of location information
G01C 21/20 - Instruments for performing navigational calculations
26.
GENERATING A SUGGESTED SHOPPING LIST BY POPULATING A TEMPLATE SHOPPING LIST OF ITEM CATEGORIES WITH ITEM TYPES AND QUANTITIES BASED ON A SET OF COLLECTION RULES
An online system generates a template shopping list for a user by accessing a machine learning model trained based on historical order information associated with the user, applying the model to predict likelihoods of conversion for item categories by the user, and populating the template shopping list with one or more item categories based on the predicted likelihoods. The system ranks one or more item types associated with each item category in the template shopping list and determines a set of collection rules associated with one or more item categories/types based on the historical order information. The system generates a suggested shopping list by populating each item category in the template shopping list with one or more item types and a quantity of each item type based on the ranking and rules and sends the suggested shopping list and rules for display to a client device associated with the user.
An online system may include a multi-agent code generator that receives webpage data describing a webpage with target content, identifies the target content by analyzing the structure of the webpage, and generates a script configured to extract the target content. The online system can execute the script to extract the target content and store the extracted data in a database for later access by the online system. For example, a chatbot of the online system can reference the stored data describing the target content to generate a response to a query.
An online system interfaces with an LLM to evaluate chatbot responses to user inputs in a conversation. The online system divides the conversation into portions and prompts the LLM to separately evaluate the chatbot's latest response in each portion. These conversation portions may include different amounts of the conversation and may build off of one another such that some portions include inputs/responses of other portions. To evaluate a chatbot's latest response in a portion, the online system may prompt the LLM to generate a score for the chatbot's response in the portion according to a conversation criterion. The prompt may instruct the LLM to consider the context of previous inputs/responses in that potion to generate the score. The online system reviews the scores and determines if any of the scores are below corresponding criteria thresholds. If so, the online system may perform a remedial action for the entire conversation.
A barcode decoding system decodes item identifiers from images of barcodes. The barcode decoding system receives an image of a barcode and rotates the image to a pre-determined orientation. The barcode decoding system also may segment the barcode image to emphasize the portions of the image that correspond to the barcode. The barcode decoding system generates a binary sequence representation of the item identifier encoded in the barcode by applying a barcode classifier model to the barcode image, and decodes the item identifier from the barcode based on the binary sequence representation.
G06K 7/14 - Methods or arrangements for sensing record carriers by electromagnetic radiation, e.g. optical sensingMethods or arrangements for sensing record carriers by corpuscular radiation using light without selection of wavelength, e.g. sensing reflected white light
A system and a method are disclosed for predicting future user engagement with a mobile device application based on a discrepancy detected between two physical objects. In an embodiment, a physical object provider receives, based on user input into the application, a request for delivery of a first physical object. A discrepancy is detected, the discrepancy reflecting that a second physical object is detected in place of the first physical object. A first set of features of the first physical object and a second set of features of the second physical object are inputted into a machine learning model. The machine learning model outputs a measure of predicted future engagement of the user with the application based on the discrepancy. The application is instructed to output an intervention based on the measure of predicted future engagement of the user.
An online system customizes documents for a particular context, user, or set of users. The online system receives an image of a physical document and extracts components, such as text, titles, items and their metadata, from the physical document. The online system may apply rules to the metadata for one or more items to determine whether to modify at least a portion of the metadata. The online system also applies a model to generate an affinity score for a context or a user and each component of the document. If the score for a component is below a threshold, the online system prompts a generative model to generate replacement content for the component. Subsequently, the online system applies the model to the generated replacement content and updates the document with the generated replacement content for the component if the score of the generated replacement content is higher.
An online concierge system generates the value for an impression by predicting future behavior by users beyond a current conversion. The predicted future behavior attributes incremental value of subsequent conversions by the user. The online concierge system gathers feature information about the user. Based on experimental data, the online concierge system generates a baseline curve describing expected user behavior for a category of users. Based on feature information of the user, the online concierge system applies a computer model to generate modifiers for the baseline curve to customize the baseline curve for the user. The modified curve is used to predict future actions by the user, and consequently a long-term incremental conversion value for the impression.
An online concierge system receives an order including one or more items from a customer and a picker obtains the items from a retailer. Upon completing obtaining items from the order and moving to checkout from the retailer, the picker updates an order status via a picker application. Via the picker application, the picker may capture an image of a shelf of items in the checkout line. The online concierge system identifies one or more items in the image using image processing and ranks the identified items for the customer from whom the order was received. The online concierge system includes a subset of the identified items ordered based on the ranking in a message to the customer via a communication interface between the customer and the picker. The message indicates the customer can add one or more of the identified items before the picker completes a checkout process.
An online concierge system selectively replaces default static item displays with dynamic item displays to represent items. The dynamic item displays encourage a viewing user of the online concierge system to purchase the items and may be selected based on item or user preferences or characteristics. The online concierge system applies a machine learning model to determine display scores describing the expected benefit of dynamic item displays and bandwidth scores describing resource usage of dynamic item displays. The online concierge system selectively replaces default static item displays with dynamic item displays based on the display and bandwidth scores so as to maximize benefit while ensuring that performance of the online concierge system is not negatively impacted by the resource usage.
A smart system, such as a smart shopping cart system, uses an efficient selection algorithm to select an item identifier prediction for an item. The smart cart system uses a set of machine-learning models to generate identifier predictions based on images. To select an item identifier, the smart system applies an efficient selection algorithm to the predictions from the machine-learning models. An efficient selection algorithm is an algorithm that requires minimal computational resources to perform. For example, the efficient selection algorithm may be a simple majority algorithm that selects the identifier prediction generated by a majority of the models or a weighted voting algorithm where each model's vote is weighted by some metric. The smart system applies the efficient selection algorithm to select an item identifier prediction from the ones generated by the models. The smart system may display content related to the item associated with the item identifier prediction.
An online system uses benchmarking tests to identify indexing algorithms for an embedding database. To perform these benchmarking tests, the online system receives a set of parameters for configuring an embedding database. For example, the parameters may include a performance parameter and a latency parameter. The online system generates algorithm scores for a set of candidate indexing algorithms based on the parameters. Specifically, the online system tests each of the candidate indexing algorithms by generating a testing database based on a subset of the entries for the full database and by performing benchmarking tests on the testing database. The online system uses these tests to compute performance metrics for each candidate indexing algorithm and uses those performance metrics to compute an algorithm score for each indexing algorithm. The online system uses the computed algorithm scores to select an indexing algorithm for the embedding database.
An online concierge system receives, from a client device associated with a user of the online concierge system, order data associated with an order placed with the online concierge system, in which the order data describes a delivery location for the order. The online concierge system receives information describing a set of attributes associated with the delivery location and accesses a machine learning model trained to predict a difference between an arrival time and a delivery time for the delivery location. The online concierge system applies the model to the set of attributes associated with the delivery location to predict the difference between the arrival time and the delivery time for the delivery location and determines an estimated delivery time for the order based at least in part on the predicted difference. The online concierge system sends the estimated delivery time for the order for display to the client device.
An online concierge system uses a machine learning click through rate model to select promoted items based on user embeddings, item embeddings, and search query embeddings. Embeddings obtained by an embedding model may be used as inputs to the click through rate model. The embedding model may be trained using different actions to score the strength of a customer interaction with an item. For example, a customer purchasing an item may be a stronger signal than a customer placing an item in a shopping cart, which in turn may be a stronger signal than a customer clicking on an item. The online concierge system generates a ranking of candidate promoted items based on the search query and using the click through rate model. Based on the ranking, the online concierge system displays promoted items along with the organic search results to the customer.
INCREMENTALLY UPDATING EMBEDDINGS FOR USE IN A MACHINE LEARNING MODEL BY ACCOUNTING FOR EFFECTS OF THE UPDATED EMBEDDINGS ON THE MACHINE LEARNING MODEL
An online concierge system uses a model to predict a user's interaction with an item, based on a user embedding for the user and an item embedding for the item. For the model to account for more recent interactions by users with items without retraining the model, the online concierge system generates updated item embeddings and updated user embeddings that account for the recent interactions by users with items. The online concierge system compares performance of the model using the updated item embeddings and the updated user embeddings relative to performance of the model using the existing item embeddings and user embeddings. If the performance of the model decreases, the online concierge system adjusts the updated user embeddings and the updated item embeddings based on the change in performance of the model. The adjusted updated user embeddings and adjusted updated item embeddings are stored for use by the model.
A system receives an order for fulfillment from a customer device, the order associated with a delivery time. The system determines a base compensation value for the order and sends the order and base compensation value to devices of one or more fulfillment agents. If the order is not accepted within a predetermined time, the system applies a trained machine learning model to updated input features of the order and the fulfillment agents to predict an amount of lateness time past the delivery time. Based on the predicted amount of lateness time, the system determines an updated lateness value, determines an updated compensation value, and sends the order with the updated compensation value to the fulfillment agents. The system repeats prediction, lateness value determination, and compensation adjustment until the order is accepted.
An online system predicts time to park at a fulfillment location in fulfillment of an order by a fulfillment user. The online system receives an order from a requesting user, and applies a timeliness prediction model to the order, the parking configuration of the corresponding fulfillment location, to other contextual factors, or some combination thereof to predict the time to park at the fulfillment location. The timeliness prediction model is trained on historical orders with their associated completion times and known parking configurations of the respective fulfillment locations. The online system may batch orders together to optimize fulfillment efficiency in consideration of the predicted lag time for the order. The online system assigns and transmits the batches to fulfillment users to fulfill at the fulfillment locations.
An online system uses benchmarking tests to identify indexing algorithms for an embedding database. To perform these benchmarking tests, the online system receives a set of parameters for configuring an embedding database. For example, the parameters may include a performance parameter and a latency parameter. The online system generates algorithm scores for a set of candidate indexing algorithms based on the parameters. Specifically, the online system tests each of the candidate indexing algorithms by generating a testing database based on a subset of the entries for the full database and by performing benchmarking tests on the testing database. The online system uses these tests to compute performance metrics for each candidate indexing algorithm and uses those performance metrics to compute an algorithm score for each indexing algorithm. The online system uses the computed algorithm scores to select an indexing algorithm for the embedding database.
An online system receives a query from a user of the online system. The online system identifies a candidate set of cold start results to the query defined as having been presented to the user less than a threshold number of times. The cold start results are then filtered based on their relevance to the query to generate a final set of cold start results and a score is generated for each cold start result without interaction data using a scoring baseline common to standard results with interaction data. Accordingly, the online system ranks the cold start results with a set of standard results based on the score for each cold start result using the scoring baseline and presents the same for display to the user.
A smart system, such as a smart shopping cart system, uses an efficient selection algorithm to select an item identifier prediction for an item. The smart cart system uses a set of machine-learning models to generate identifier predictions based on images. To select an item identifier, the smart system applies an efficient selection algorithm to the predictions from the machine-learning models. An efficient selection algorithm is an algorithm that requires minimal computational resources to perform. For example, the efficient selection algorithm may be a simple majority algorithm that selects the identifier prediction generated by a majority of the models or a weighted voting algorithm where each model's vote is weighted by some metric. The smart system applies the efficient selection algorithm to select an item identifier prediction from the ones generated by the models. The smart system may display content related to the item associated with the item identifier prediction.
An online system uses a trained machine-learning model to predict hard-to-find items, which may facilitate picking of these items. The online system receives, from one or more devices of one or more pickers, a device of a source, one or more devices associated with one or more users, and/or a computing system associated with a physical receptacle utilized by at least one user for shopping in a location of the source, data with information about an item. The online system applies the trained machine-learning model to output, based on the received data, a findability score for the item indicative of a findability of the item. Based on the findability score, the online system generates and communicates one or more action signals to a device of a picker, the device of the source, and/or a device associated with a user prompting one or more actions in relation to the item.
G06Q 10/087 - Inventory or stock management, e.g. order filling, procurement or balancing against orders
G06Q 30/02 - MarketingPrice estimation or determinationFundraising
G06V 10/77 - Processing image or video features in feature spacesArrangements for image or video recognition or understanding using pattern recognition or machine learning using data integration or data reduction, e.g. principal component analysis [PCA] or independent component analysis [ICA] or self-organising maps [SOM]Blind source separation
46.
Display panel of a programmed computer system with a graphical user interface
An online system displays an interface to users including slots in which sources from a list of sources of items (e.g., physical items, content items) are presented. The user may select a source via the interface to view items provided by, or associated with, the source. To simplify a user identifying a desired source, the online system includes sources that a user is likely to select as well as new sources in the list. To balance the competing interests of relevance of sources with which the user previously interacted and discovery of new sources, the online system selects an allocation of slots for new sources and for sources with prior interaction based on interactions by users in the geographic regions with different allocations of slots. Based on the selected allocation of slots, the online system selects specific retailers for each slot using ranking models corresponding to different slots.
A client device or an online system communicating with the device receives video data captured by a camera of the device, in which the video data depicts a field of view of a display area of the device. The device/ system detects an object within the field of view based on the video data and applies one or more machine-learning algorithms to identify the object as an item available at a source. The device/system accesses item data for items available at the source and selects one or more supplemental items associated with the identified item based on item data for the identified item and each supplemental item. The device/system generates an augmented reality element including a listing of the supplemental item(s). as well as information or a selectable option associated with each supplemental item. The augmented reality element is then displayed in the display area of the device.
G06T 19/00 - Manipulating 3D models or images for computer graphics
G06V 10/22 - Image preprocessing by selection of a specific region containing or referencing a patternLocating or processing of specific regions to guide the detection or recognition
G06V 10/72 - Data preparation, e.g. statistical preprocessing of image or video features
G06V 10/94 - Hardware or software architectures specially adapted for image or video understanding
49.
Using a Trained Machine-Learning Model of an Online System to Handle Unclaimed Online Pickup Orders
An online system uses a trained model for intelligent handling of unclaimed online pickup orders. After identifying that an order placed by a user of the online system is unclaimed at a location of a source, the online system obtains, from a device of a picker associated with the online system and/or a device associated with the source, signals with information about each item in each bundle of the unclaimed order. The online system applies the trained model to identify, based on the obtained signals, a preferred method for disposal of each bundle. Based on the identified preferred method for disposal of each bundle, the online system generates a disposal decision signal and communicates the disposal decision signal to the device associated with the source that prompts personnel at the location of the source to dispose each bundle of the unclaimed order using the identified preferred disposal method.
An online concierge system generates suggested instructions for presentation to a user. The online concierge system access instruction examples corresponding to a target item category and generates candidate instruction representations based on instruction messages within each instruction example. The online concierge system generates preliminary scores for the candidate instruction representations that are directly related to an intra-category frequency of use of the instruction tokens of the candidate instruction representation within the target item category. The online system normalizes these preliminary scores for the candidate instruction representations based on the inter-category frequency of use of the instruction tokens in all item categories to generate final scores for the candidate instruction representations. The online concierge system selects a set of instruction representations based on these final scores and generates suggested instructions based on the set of instruction representations.
An online system automatically identifies item attributes of an item. The online system prompts a set of outputs from a set of multi-modal large language models with an image of the product and a request to determine if the details of size information is present in the image. The online system receives a set of outputs, wherein an output describes whether the size information is present in the image. The system then prompts the set of language models with a request to extract the value of the size information in the image. Responsive to determining that a threshold number of outputs have matching values of size information that is present in the image, the system updates the item attribute data with the matching values of size information of the product.
An online system receives, at a search interface, a search query from a user. The online system determines a recall set size for search results of the search query. The online system determines a page load size to display at least a portion of the search results by determining a query entropy associated with the search query, inputting a plurality of signals into a machine learning model, the plurality of signals comprising the query entropy, and receiving, from the machine learning model, the page load size. The online system selects a set of physical object identifiers based on the page load size. The online system generates for display a user interface that groups the selected physical object identifiers. The online system causes a device associated with the user to display the generated user interface.
An online system uses a trained model to predict incremental sales caused by a sample counter for in-store free sampling of an item. Upon receiving signals related to in-store purchases of the item, the online system applies the trained model to output, based on the received signals, a ranked list of locations of a source and a ranked list of timeslots for placing the sample counter. The online system selects, from the ranked list of locations and the ranked list of timeslots, a location of the source and a timeslot for placing the sample counter, and generates a decision signal based on the selected location and the selected timeslot. The online system communicates, via the network to a device associated with the source, the decision signal prompting the source to place the sample counter for free sampling of the item at the selected location and during the selected timeslot.
A client device or an online system communicating with the device receives video data captured by a camera of the device, in which the video data depicts a field of view of a display area of the device. The device/system detects an object within the field of view based on the video data and applies one or more machine-learning algorithms to identify the object as an item available at a source. The device/system accesses item data for items available at the source and selects one or more supplemental items associated with the identified item based on item data for the identified item and each supplemental item. The device/system generates an augmented reality element including a listing of the supplemental item(s), as well as information or a selectable option associated with each supplemental item. The augmented reality element is then displayed in the display area of the device.
A client device, or an online system communicating with the device, receives video data depicting a field of view of a display area of the device and applies machine-learning algorithms to the video data to detect objects, including portions of a body of a user of the device, within the field of view and to determine a series of body poses. The device/system uses machine-learning models to predict an action performed by the user based on the series of poses and to predict a recipe being prepared based on the objects and a predicted series of actions performed by the user. The device/system selects a suggestion associated with preparing the recipe based on candidate suggestions associated with preparing the recipe, the objects, or the predicted series of actions, and generates an augmented reality element describing the suggestion. The augmented reality element is displayed in the display area of the device.
An online system retrieves a set of user data including information describing one or more interactions by a user with the system. The system accesses and applies a machine-learning model to predict an exploration score for the user based on the set of user data, in which the score describes a likelihood of a set of interactions by the user with content associated with less than a threshold measure of familiarity to the user. Upon receiving a request from a client device associated with the user to access a user interface including content recommended to the user, the system selects content to recommend to the user based on the score and information describing a set of previous interactions by the user with the content. The system generates the user interface including the selected content and sends the user interface to the client device where it is displayed.
A trained model is used to determine a price sensitivity feature for a user of an online system. The online system generates input data by gathering replacement data via a user interface at a device associated with the user and/or in-store behavior data related to replacement of items performed by the user at a location of a retailer when using a physical receptacle in communication with the online system. The online system applies a price sensitivity model to predict, based on the input data, a price sensitivity score for the user indicative of the price sensitivity feature of the user. The online system identifies, based on the price sensitivity score, one or more actions related to prompting the user to convert one or more items. The online system applies the one or more actions to prompt the user to convert the one or more items.
Based on orders fulfilled by shoppers of an online concierge system, the online concierge system identifies items in an order that are difficult to find in a warehouse in which the order is fulfilled. When a shopper obtains a difficult to find item from the warehouse, the online concierge system prompts the shopper to provide information for finding the difficult to find item in the warehouse. The online concierge system stores the information for finding the difficult to find item from the shopper in association with the difficult to find item and with the warehouse. Subsequently, when a different shopper is fulfilling an order from the warehouse including the difficult to find item, the online concierge system displays the information for finding the difficult to find item in the warehouse to the different shopper.
A device may obtain historical order data comprising orders submitted by users to an online system, each order indicating a retailer location and a timestamp. A device may generate a first set of training examples, each training example indicating order demand at a retailer location during one period of time from a first set of periods of time. A device may train the demand forecast prediction model with the first set of training examples. A device may apply the demand forecast prediction model to a second set of periods of time to predict order demand for each period of time in the second set of periods of time. A device may track order demand across each period of time in the second set of periods of time. A device may generate a second set of training examples, each training example indicating a difference between the predicted order demand and the tracked order demand at the retailer location during each period of time from the second set of periods of time. A device may retrain the demand forecast prediction model with the second set of training examples.
A trained model is used to generate a user interface of an online system based on predicted nutritional preferences for a user of the online system. Upon receiving a signal indicating interaction of the user with the online system, the online system applies the trained model to output, based on user's features, item features and/or session features, a vector of scores for the user, where each score is indicative of a preference of the user for a respective nutritional attribute of a set of nutritional attributes. Responsive to a score being greater than a threshold score, the online system generates, based on the received signal, a user interface of a device associated with the user that includes a label about the nutritional attribute associated with the score. The online system causes the device associated with the user to display the user interface with the label about the nutritional attribute.
A user interface of an online system is generated based on search for relevant items that match ingredients of a recipe. After receiving, from a device associated with a user of the online system, a query for an ingredient of a recipe, the online system identifies, based on one or more attributes in the query, a set of candidate items for the ingredient. The online system generates a recipe relevance score for each candidate item by applying a weighted sum of scores, ranks the identified candidate items based on their recipe relevance scores, and selects one or more items for presentation to the user. The online system then generates a user interface of the device with a recipe page including the ingredient of the recipe and the one or more items that match the ingredient of the recipe.
An online concierge system fulfills orders placed by users. When a user notifies the online concierge system of a problem with order fulfillment, the online concierge system performs one or more remedial actions (e.g., a credit, a discount, a free delivery). To provide a proactive remedial action before receiving a notification of a problem with order fulfillment from a user, the online concierge system trains a proactive remediation model that predicts, for an order having an event during fulfillment, a likelihood of loss of interaction by the user with the online concierge system (i.e., “churn” of the user) without performing a proactive remedial action. When fulfilling an order, in response to determining an event during fulfillment, the online concierge system applies the proactive remediation model to determine the likelihood of churn of the user if no proactive appeasement is performed, for determining whether to perform a proactive remedial action.
An online system displays an interface to users including slots in which sources from a list of sources of items (e.g., physical items, content items) are presented. The user may select a source via the interface to view items provided by, or associated with, the source. To simplify a user identifying a desired source, the online system includes sources that a user is likely to select as well as new sources in the list. To balance the competing interests of relevance of sources with which the user previously interacted and discovery of new sources, the online system selects an allocation of slots for new sources and for sources with prior interaction based on interactions by users in the geographic regions with different allocations of slots. Based on the selected allocation of slots, the online system selects specific retailers for each slot using ranking models corresponding to different slots.
Personalized recommendations matching a list of item descriptors to catalog products from is described. A list associated with a user is received that includes item descriptors. The item descriptors correspond to catalog products stored in a catalog database that includes a plurality of catalog products. Linking data for the user is retrieved. For at least one of the item descriptors in the list, a model is applied to the linking data to generate a score for each of a set of candidate catalog products. A list of recommended catalog products for the user is built by, for each of the item descriptors in the list, selecting one of the set of candidate catalog products based on the generated scores. The list of recommended catalog products is provided to a user client device associated with the user. The user client device is configured to display the list of recommended catalog products.
An online concierge system receives, from a client device associated with a user, a request to access a user interface including a listing of sources associated with the system, in which each source is associated with a catalog of items. The system retrieves user data describing interactions by the user with items available at the sources and accesses and applies a machine-learning model to predict a user engagement score for each item-source pair associated with the sources based on the user data, in which the score indicates a likelihood of an interaction by the user with an item available at a source. The system selects a set of item-source pairs based on the scores and generates the user interface including the listing and a selectable option to add an item associated with each selected pair to a shopping list. The system then sends the user interface to the client device.
A computer system uses clustering and a large language model (LLM) to normalize attribute tuples for items stored in a database of an online system. The online system collects attribute tuples, each attribute tuple comprising an attribute type and an attribute value for an item. The online system initially clusters the attribute tuples into a first plurality of clusters. The online system generates prompts for input into the LLM, each prompt including a subset of attribute tuples grouped into a respective cluster of the first plurality. Based on the prompts, the LLM generates a second plurality of clusters, each cluster including one or more attribute tuples that have a common attribute type and a common attribute value. The online system maps each attribute tuple to a respective normalized attribute tuple associated with each cluster. The online system rewrites each attribute tuple in the database to a corresponding normalized attribute tuple.
Clustering database items based on output of machine-learning model to link database items that represent the same core item but with a different stored size or form attribute
A trained machine-learning model is used to group items in a database of an online system that represent a same core item (i.e., product) but of different attributes. The online system applies, for pairs of items from a specific chunk of the database, the machine-learning model to metadata for each pair of items and category data for each pair of items to generate a clustering score for each pair of items that indicates a likelihood that both items belong to a cluster of items that identifies the core item. The online system then requests a language model to generate a response including a list of attributes in a structured form for each item from the cluster. The online system stores, in an entry of the database associated with each item from the cluster, the list of attributes in the structured form and an identification of the cluster.
An online system performs an inference task in conjunction with the model serving system infer seasonality of items in an item catalog hosted by the online system. The online system generates and provides a prompt to a machine-learned language model to output a list of item categories predicted to be in season for a particular time period and a particular geographical location, e.g., associated with a requesting user. The language model outputs the list of item categories predicted to be in season. The online system validates the list by leveraging the language model and/or historical user engagement data. The online system maps items in the item catalog to the seasonal item categories and tags the mapped items with an in-season badge for display with the item in an ordering interface to the requesting user.
A trained model is used to predict a type of a user of an online system to generate a personalized user interface of the online system. Upon receiving data related to a current session of the user with the online system, the online system applies the trained model to output, based on the session data, a score for the user indicative of a predicted type of the user for the current session. The online system compares the score with a threshold score, and responsive to the score being greater than the threshold score, the online system identifies, based on the score, user data, and information about the current session, a set of user interface elements arranged in a specific order for presentation to the user. The online system then generates a user interface of the device associated with the user that includes the arranged user interface elements.
A machine learned model for item recommendations following failed attempts to purchase those items. During a session, an online system receives a request to fulfill an order from a user device. The system receives a message indicating that an item from the order was not fulfilled. The system logs the item in connection with a profile of the user stored in a database of the online system. During a subsequent session with the user device, the system determines that the logged item is available for fulfillment. The system applies the model to output an intent score indicative of an intent of a user of the user device to acquire the logged item. The logged item is ranked based on the intent score, and a user interface is generated that includes a recommendation to acquire the logged item. The system causes the user device to display the generated user interface.
An online system receives a user query for execution by at least one of a set of generative artificial intelligence (AI) models. The online system assigns the user query to one or more query categories of a set of query categories. The online system accesses a dataset stored in a database. For each query category, the dataset stores a preferred generative AI model for the query category among the set of generative AI models. The online system selects a preferred generative AI model for the user query from the database based on the one or more query categories assigned to the user query. The online system provides a prompt to a model serving system hosting the selected generative AI model. The online system receives, from the model serving system, a response to the user query generated by executing the prompt.
An online concierge system may use images received from shopping carts within retailers to determine the availability of items within those retailers. A shopping cart includes externally-facing cameras that automatically capture images of the area around the shopping cart as the shopping cart travels through a retailer. The online concierge system receives these images, which depict displays within the retailers from which a picker or a retailer patron can collect items. The online concierge system determines which items should be depicted in the images and which items are actually depicted in the images. The online concierge system identifies which items should be depicted, but are not depicted, and determines that these items are unavailable (e.g., out of stock) at that retailer. The online concierge system updates an availability database to indicate that these items are unavailable and may notify the retailer that the item is unavailable.
A machine-learned predictive model is trained to predict potential for customer complaint. The model is part of an online concierge system. The online concierge system accesses a customer order that includes one or more items. The online concierge system determines input data for an item of the one or more items. The online concierge system determines a prediction value associated with potential for customer complaint for the item by applying the machine-learned prediction model to the input data. The online concierge system provides the prediction value to a picker client device associated with a picker who is assigned the item. The picker client device presents an alert to the picker based in part on the prediction value, and the alert includes a message that is customized to mitigate a cause of potential customer complaint for the item.
An automated checkout system modifies received images of machine-readable labels to improve the performance of a label detection model that the system uses to decode item identifiers encoded in the machine-readable labels. For example, the automated checkout system may transform subregions of an image of a machine-readable label to adjust for distortions in the image's depiction of the machine-readable label. Similarly, the automated checkout system may identify readable regions within received images of machine-readable labels and apply a label detection model to those readable regions. By modifying received images of machine-readable labels, these techniques improve on existing computer-vision technologies by allowing for the effective decoding of machine-readable labels based on real-world images using relatively clean training data.
G06K 7/14 - Methods or arrangements for sensing record carriers by electromagnetic radiation, e.g. optical sensingMethods or arrangements for sensing record carriers by corpuscular radiation using light without selection of wavelength, e.g. sensing reflected white light
An online system trains a machine-learned lift prediction model configured as a neural network. The machine-learned lift prediction model can be used during the inference process to determine lift predictions for users and items associated with the online system. By configuring the lift prediction model as a neural network, the lift prediction model can capture and process information from users and items in various formats and more flexibly model users and items compared to existing methods. Moreover, the lift prediction model includes at least a first portion for generating control predictions and a second portion for generating treatment predictions, where the first portion and the second portion share a subset of parameters. The shared subset of parameters can capture information important for generating both control and treatment predictions even when the training data for a control group of users might be significantly smaller than that of the treatment group.
An online system receives a user query for execution by at least one of a set of generative artificial intelligence (Al) models. The online system assigns the user query to one or more query categories of a set of query categories. The online system accesses a dataset stored in a database. For each query category, the dataset stores a preferred generative Al model for the query category among the set of generative Al models. The online system selects a preferred generative Al model for the user query from the database based on the one or more query categories assigned to the user query. The online system provides a prompt to a model serving system hosting the selected generative Al model. The online system receives, from the model serving system, a response to the user query generated by executing the prompt.
An online concierge system maintains a graph of items available for purchase. The graph maintains edges between items, where an edge between an item and an additional item indicates that one or more customers have previously replaced the item with the additional item. The edge between the item and the additional item also identifies a number of times customers have replaced the item with the additional item. When a customer orders an item, the online concierge system traverses the graph of items to identify candidate replacement items for the ordered item and identifies one or more of the candidate replacement items to the customer. When identifying the candidate replacement items, the online concierge system accounts for distance between the ordered item and different candidate replacement items in the item graph.
A trained model detects seasonal items in an item catalog database of an online system. Upon acquiring item data with information about an item in the item catalog database, the online system applies the trained model to output, based on the item data, a seasonality score for the item that is indicative of a predicted seasonality of the item, and to identify a season associated with the item. The online system updates the item catalog database by adding an identification of the identified season to an entry of the item in the item catalog database. The online system further generates, based on the seasonality score and the identified season, action data associated with one or more actions in relation to the item. The online system communicates, to a computing system of a retailer, the action data prompting the one or more actions in relation to the item.
An online system uses a computer-vision item identification model to identify items and physical containers storing those items to detect sorting errors of the physical containers. The online system receives a first image from a client device that depicts a set of physical containers that contain items for a batch of orders that the online system has received. The online system identifies items in those physical containers by applying a contained-item identification model to the first image. The online system uses the output of this model to determine which visible items are in each physical container and uses that information plus order data for the batch of orders to determine which physical containers are associated with each order. The online system compares this first image to a subsequently received image to determine whether the correct physical containers were delivered by the user.
An online system predicts whether a user will be at a delivery location at a delivery time for an attended delivery of an order using a machine-learned model. The online system receives the order from a client device of a user and a request by the user for an attended delivery of the order where the user will be at the delivery location at the delivery time of the order. The machine-learned model predicts that the user will not be at the delivery location at the delivery time based on user attributes of the user and order attributes of the order that are input into the machine-learned model. The online system performs a remedial action including transmitting a notification to the client device of the user to provide additional instructions for the attended delivery responsive to the determination that the user is not likely to be at the delivery location.
An online concierge system receives item data for an item included among an inventory at a retailer location, in which the item data includes a set of real-time item data for the item and a set of constraints. The system accesses and applies a first machine-learning model to predict a freshness satisfaction score for the item based at least in part on the item data. The system updates the item data to include the score and accesses and applies a second machine-learning model to predict an elasticity of demand for the item based at least in part on the updated item data. The system determines an optimal value associated with the item based at least in part on the freshness satisfaction score, the elasticity of demand, and the set of constraints. A value associated with the item is then adjusted based at least in part on the optimal value.
An online system configures one or more system Al agent instances that interact with user Al agents and performs one or more tasks on behalf of the online system. Thus, responsive to detecting the presence of a user Al agent representing a particular user, the online system directs the session for the user to communicate and interact with a system Al agent.
An online system configures one or more system AI agent instances that interact with user AI agents and performs one or more tasks on behalf of the online system. Thus, responsive to detecting the presence of a user AI agent representing a particular user, the online system directs the session for the user to communicate and interact with a system AI agent.
A chat interface supported by language models is used for generating a group order at an online system based on a conversation between multiple users. Upon receiving, via the chat interface, input data with information about the conversation, the online system requests a first language model to generate, based on the input data, a list of ingredients. The online system then requests a second language model to map the list of ingredients into a list of items at a retailer associated with the online system. Upon generation of the list of items, the online system causes the chat interface to display content prompting approval by the users for conversion of the list of items. Responsive to the approval, the online system places the group order that includes the list of items for delivery to a user of the online system.
G06F 40/40 - Processing or translation of natural language
H04L 51/02 - User-to-user messaging in packet-switching networks, transmitted according to store-and-forward or real-time protocols, e.g. e-mail using automatic reactions or user delegation, e.g. automatic replies or chatbot-generated messages
86.
PERSONALIZED PRESENTATION OF CONTENT BASED ON LOCATION DATA CAPTURED FROM SMART CART SYSTEMS
A system stores content items at a content data store, each content item corresponding to an item within an environment. The system accesses location data captured by a plurality of location sensors in the environment, each coupled to a smart cart system located within the environment. The location data indicates a current location of a corresponding smart cart system. The system computes a number of smart cart systems within a threshold area around a display screen within the environment based on the location data. The system computes a presentation score for each of the content items by combining a context relevance score and a personal relevance score weighted based on a personalization weighting. The system selects a subset of the content items for display based on the presentation scores and causes the display screen to present the subset of content items.
A system causes a display screen to present a set of content for a first time period. The system accesses a first set of location data captured by location sensors coupled to shopping carts. The first set of location data indicates a location of each of a plurality of users of the shopping carts ]. The system identifies a set of users within a distal proximity of the display screen during the first time period. The system accesses sensor data captured by sensors of the shopping carts and detects an action performed by a first user in relation to a first item. The system identifies a timestamp for the action. In response to the timestamp being within a threshold amount of time after the first time period, the system stores proximity data indicative of an interaction with the set of content in association with the first user.
A system may store a plurality of images depicting items within an environment, where each image was captured by a camera coupled to a shopping cart. The system identifies a target item associated with a user device that is located within the environment. The system identifies a location of the target item within the environment based on item data associated with the target item and environment map data describing the environment. The system selects, from the plurality of images, an image depicting the item at the location within the environment based on the environment map data and the location data associated with each of the plurality of images. The system identifies a portion of the identified image that depicts the item by applying a machine-learning model to the identified image. The system modifies the identified portion of the identified image to highlight the target item.
A smart shopping cart identifies items using cameras and sensors. The cart captures images of items within its storage area and applies machine-learning models, such as a barcode detection model, an OCR model, and an image embedding model, to generate identifier predictions. These predictions are processed using an efficient selection algorithm, which may involve majority voting, weighted voting, or linear regression, to select the most accurate identifier. The cart updates its display and user interface with the identified item. The process may be performed primarily by the CPU to enhance computational efficiency, avoiding the latency associated with GPU data transfer. Additional techniques, such as circular buffers and frame skipping, are employed to further optimize resource usage.
G06V 10/70 - Arrangements for image or video recognition or understanding using pattern recognition or machine learning
G06V 10/764 - Arrangements for image or video recognition or understanding using pattern recognition or machine learning using classification, e.g. of video objects
A system stores content items at a content data store, each content item corresponding to an item within an environment. The system accesses location data captured by a plurality of location sensors in the environment, each coupled to a smart cart system located within the environment. The location data indicates a current location of a corresponding smart cart system. The system computes a number of smart cart systems within a threshold area around a display screen within the environment based on the location data. The system computes a presentation score for each of the content items by combining a context relevance score and a personal relevance score weighted based on a personalization weighting. The system selects a subset of the content items for display based on the presentation scores and causes the display screen to present the subset of content items.
A system causes a display screen to present a set of content for a first time period. The system accesses a first set of location data captured by location sensors coupled to shopping carts. The first set of location data indicates a location of each of a plurality of users of the shopping carts]. The system identifies a set of users within a distal proximity of the display screen during the first time period. The system accesses sensor data captured by sensors of the shopping carts and detects an action performed by a first user in relation to a first item. The system identifies a timestamp for the action. In response to the timestamp being within a threshold amount of time after the first time period, the system stores proximity data indicative of an interaction with the set of content in association with the first user.
A system may store a plurality of images depicting items within an environment, where each image was captured by a camera coupled to a shopping cart. The system identifies a target item associated with a user device that is located within the environment. The system identifies a location of the target item within the environment based on item data associated with the target item and environment map data describing the environment. The system selects, from the plurality of images, an image depicting the item at the location within the environment based on the environment map data and the location data associated with each of the plurality of images. The system identifies a portion of the identified image that depicts the item by applying a machine-learning model to the identified image. The system modifies the identified portion of the identified image to highlight the target item.
An online system may generate numerous search records in response to searches requested by users. The online system may use a specific way to sample the historical search records to reduce biases in sampling. For example, the online system retrieves historical query records associated with an item query engine. The set of historical query records includes a plurality of search phrases. A historical query record is associated with a search phrase and a list of items returned by the item query engine. The online system determines the search frequencies for the search phrases. The online system stratifies the historical query records into a plurality of bins according to the search frequencies of the search phrases. The online system samples the historical query records from the plurality of bins to collect a representative set of historical query records and outputs the representative set of historical query records for rating.
A smart shopping cart identifies items using cameras and sensors. The cart captures images of items within its storage area and applies machine-learning models, such as a barcode detection model, an OCR model, and an image embedding model, to generate identifier predictions. These predictions are processed using an efficient selection algorithm, which may involve majority voting, weighted voting, or linear regression, to select the most accurate identifier. The cart updates its display and user interface with the identified item. The process may be performed primarily by the CPU to enhance computational efficiency, avoiding the latency associated with GPU data transfer. Additional techniques, such as circular buffers and frame skipping, are employed to further optimize resource usage.
An online security system identifies matching client devices by comparing location data points acquired during application workflow from different devices. The location data points comprise a location of the device and a timestamp. Identifiable pairs of data points are collated from different devices when they show a device present within a threshold distance or the same geographic region at identical times. The system utilizes a set of matching criteria to decide whether one data set for one device corresponds with a set from another. Verification of the matches allows the system to ascertain the same user is operating both devices and link the user to both devices. This system enhances security by identifying users likely gaining unauthorized access through multiple device usage simultaneously.
METHOD, COMPUTER PROGRAM PRODUCT, AND SYSTEM FOR TRAINING A MACHINE LEARNING MODEL TO GENERATE USER EMBEDDINGS AND RECIPE EMBEDDINGS IN A COMMON LATENT SPACE FOR RECOMMENDING ONE OR MORE RECIPES TO A USER
An online concierge system generates recipe embeddings for recipes including multiple items and user embeddings for users, with the recipe embeddings and user embeddings in a common latent space. To generate the user embeddings and the recipe embeddings, a model includes separate layers for a user model outputting user embeddings and for a recipe model outputting recipe embeddings. When training the model, a weight matrix generates a predicted dietary preference type for a user embedding and for a recipe embedding and adjusts the user model or the recipe model based on differences between the predicted dietary preference type and a dietary preference type applied to the user embedding and to the recipe embedding. Additionally cross-modal layers generate a predicted user embedding from a recipe embedding and generate a predicted recipe embedding from a user embedding that are used to further refine the user model and the recipe model.
A scrollable listing of icons associated with catalogs is displayed, in which the scrollable listing of icons is overlaid onto a page when the page is scrolled, each icon is associated with a catalog, and each icon is displayed with an indication of a set of items selected from a corresponding catalog. Status information may be displayed with one or more of the icons. In response to a user selection of an icon from the scrollable listing of icons, the page is updated to include items included in a catalog associated with the selected icon. In response to a user selection to add an item from the page including the items, the indication displayed with the selected icon in the scrollable listing of icons is updated to indicate the added item.
An online concierge system performs asynchronous automated correction handling of incorrectly sorted items using point-of-sale data. The online concierge system receives orders from customer client devices and determines a batched order based on the received orders. The online concierge system sends the batched order to a shopper client device for fulfillment. The online concierge system receives transaction data associated with the batched order from a third party system. The online concierge system determines whether a sorting error occurred based on the transaction data and the batched order. In response to determining that a sorting error occurred, the online concierge system sends an instruction to correct the sorting error to the shopper client device.
G06Q 10/087 - Inventory or stock management, e.g. order filling, procurement or balancing against orders
G06Q 20/40 - Authorisation, e.g. identification of payer or payee, verification of customer or shop credentialsReview and approval of payers, e.g. check of credit lines or negative lists
99.
MACHINE LEARNING MODEL FOR DETERMINING A TIME INTERVAL TO DELAY BATCHING DECISION FOR AN ORDER RECEIVED BY AN ONLINE CONCIERGE SYSTEM TO COMBINE ORDERS WHILE MINIMIZING PROBABILITY OF LATE FULFILLMENT
An online concierge identifies orders to shoppers, allowing shoppers to select orders for fulfillment. The online concierge system may generate batches that include multiple orders, allowing a shopper to select a batch to fulfill multiple orders. As orders are continuously being received, delaying identification of orders to shoppers may allow greater batching of orders. To allow greater opportunities for batching, the online concierge system estimates a benefit for delaying identification of an order by different time intervals and predicts an amount of time to fulfill the order. The online concierge system then delays assigning orders for which there is a threshold benefit for delaying and selects a time interval for delaying identification of the order that does not result in greater than a threshold likelihood of a late fulfillment of the order.
ALLOCATING SHOPPERS FOR ORDER FULFILLMENT BY AN ONLINE CONCIERGE SYSTEM ACCOUNTING FOR VARIABLE NUMBERS OF SHOPPERS ACROSS DIFFERENT TIME WINDOWS AND VARYING CAPABILITIES FOR FULFILLING ORDERS
An online concierge system facilitates order fulfillment by maintaining discrete time intervals for deliveries and a hierarchical data structure encoding picker characteristics. Each level in the tree structure represents a fulfillment capability and is assigned a value. Upon receiving an order specifying items and a time interval, the system applies a machine learning model to predict the number and capability levels of available pickers for the specified interval. The model is trained using historical data labeled with picker availability and their corresponding capability levels. Training includes predicting picker counts per level, computing an error metric, and updating model parameters to minimize error. The system further analyzes the order to assign tags that map to required picker characteristics, aligning them with corresponding levels in the tree structure.
