Abstract: One or more aspects of the method, apparatus, and non-transitory computer readable medium include receiving a query relating to an item and a summarization type indicating an emphasis on item similarities or item differences, obtaining, using a search component, descriptions of items relevant to the query, generating input data for a machine learning model based on the descriptions and the summarization type, and generating, using the machine learning model, a summarization of the descriptions based on the input data in response to the query, wherein the summarization emphasizes the item similarities or item differences based on the summarization type.

Abstract: In accordance with the described techniques, a tone embedding system includes one or more generative content models and a neural network trained to extract tone from input content. The tone embedding system receives a set of multimodal input content associated with an entity, and a content generation prompt. Based on the content generation prompt, the one or more generative content models produce generated content. Further, the neural network is employed to generate an entity embedding capturing a tone conveyed by the set of multimodal input content, as well as a content embedding capturing a tone conveyed by the generated content. The generated content is output based on a degree of alignment between the entity embedding and the content embedding meeting a tone alignment threshold.

Abstract: Aspects of the technology described herein relate to performing item retrieval using generative inpainting. In accordance with some aspects, an image having one or more objects is accessed. Responsive to user input, a mask is applied to the image to provide a masked image, where the mask overlays at least a portion of a first object from the one or more objects. A generative model generates an inpainted image by inpainting the mask of the masked image. One or more search results are identified using the inpainted image and are provided for presentation.

Abstract: Systems and methods for detecting object demands based on query images are provided. An image tagging module generates multiple image tags for a query image. An image content analyzer module analyzes the multiple query image tags based on a knowledge graph associated with an online platform to create query feature data. A theme identification module identifies one or more query themes based on aggregated query image feature data. A demand analysis module generates demand data indicating user demand for an object corresponding to the query theme by comparing the query theme to catalog data of the online platform.

Abstract: In accordance with the described techniques, a background generation system receives one or more images depicting an object, and textual information describing the object. A generative text model is employed to generate a prompt based on the one or more images and the textual information. Further, a generative image model is employed to generate an output image. To do so, the generative image model generates a background image based on the prompt, and the object is incorporated into the background image. Using a visual saliency model, the background generation system determines a visual saliency defining a degree of fixation on the object within the output image. The background generation system outputs the output image based on the visual saliency meeting a threshold.

Abstract: Systems and methods for generating geolocation-based images for a target object are provided. A geolocation module receives a set of geolocations associated with a geographic region of interest. Each geolocation of the set of geolocations is mapped to context data associated with the geolocation. A prompt generation module generates multiple prompts based on the set of geolocations and the context data. The prompt generation module comprises a first generative artificial intelligence (AI) model. An image generation module generates multiple synthetic images based on the multiple prompts. The image generation module comprises a second generative AI model. Each synthetic image depicts the target object in a background generated based on a prompt.

Abstract: Certain aspects and features of this disclosure relate to chromatic undertone detection. For example, a method involves receiving an image file and producing, using a color warmth classifier, an image warmth profile from the image file. The method further involves applying a surface-image-trained machine-learning model to the image warmth profile to produce an inferred undertone value for the image file. The method further involves comparing, using a recommendation module, and the inferred undertone value, an image color value to a plurality of pre-existing color values corresponding to a database of production images, and causing, in response to the comparing, interactive content including the at least one production image selection from the database of production images to be provided on a recipient device.

Abstract: Methods and systems disclosed herein relate generally to increasing visibility of pixel patterns of an image. The system includes a pattern-detection application accessing an image depicting an object. The pattern-detection application determines a set of colors from the transformed image. The pattern-detection application identifies a set of pixels depicting a particular color of the set of colors. For the set of pixels depicting the particular color, the pattern-detection application converts an initial set of pixel values of the set of pixels at an initial color space to another set of pixel values that define the particular color of the set of pixels in another color space. The pattern-detection application modifies one or more values of the other set of pixel values to generate a modified set of pixel values. The modification includes causing the set of pixels visually indicate a simulated color that is different from the particular color.

Abstract: Methods and systems are provided for using machine learning to optimize UoM representations. In embodiments described herein, units of measure (UoMs) and relationships of each of UoMs to textual representations of each of the UoMs are stored in a knowledge graph. Text corresponding to a measurement of a product is extracted by an inference model. A recommended textual representation of the measurement of the product by is determined by an autoencoder model including a corresponding textual representation of one of the UoMs from the textual representations of the one of the UoMs stored in the knowledge graph. The recommended textual representation of the measurement of the product is then displayed.

Abstract: An effective stock keeping unit (SKU) management system encodes catalog data into an embedding per catalog item. An embedding space is created by encoding catalog item data into an embedding per catalog item. The embedding is created by generating an index, where a number of rows represents a number of catalog items and a number of columns represents a number of fields associated with each catalog item. The index is then denormalized using customer groups and transformed by compressing the number of columns, to create the embedding space. In some configuration, a machine learning model is trained using catalog data. In the embedding space, item similarity is encoded by clustering catalog SKUs into groups in the embedding space, by placing similarly related items close to each other in the embedding space. Catalog items are then searched for in the embedding, with the closest clusters searched for a particular catalog item.

Abstract: Methods and systems are provided for using generative AI to optimize product search queries. In embodiments described herein, product descriptions and product images for a plurality of products are obtained. A multi-modal style classification model classifies each product into a corresponding style of a plurality of styles based on the product's product description and product image. Relationships of each product to other products in the plurality of products are stored in a knowledge graph based on the corresponding style of each product and the corresponding product description of each product. An image is generated by a text-to-image diffusion model with a set of products of the plurality of products based on the relationships of each product of the plurality of products to other products in the plurality of products.

Abstract: A feedback management subsystem receives, from a first user, first text comprising commentary on an item. The feedback management subsystem receives, from the first user, instructions to request commentary on the item from a second user. Responsive to receiving the instructions to request commentary from the second user, a communication subsystem transmits a notification to the second user. The feedback management subsystem receives, from the second user, second text comprising commentary on the item. A first machine learning model performs sentiment analysis to identify sentiments of the first text and the second text. A recommendation subsystem identifies prior actions of the first user and associated sentiments of the second user. A second machine learning model identifies a second item based on the prior actions of the first user and the sentiments of the second user. The recommendation subsystem provides output to the first user recommending the second item.

Abstract: An effective stock keeping unit (SKU) management system encodes catalog data into an embedding per catalog item. An embedding space is created by encoding catalog item data into an embedding per catalog item. The embedding is created by generating an index, where a number of rows represents a number of catalog items and a number of columns represents a number of fields associated with each catalog item. The index is then denormalized using customer groups and transformed by compressing the number of columns, to create the embedding space. In some configuration, a machine learning model is trained using catalog data. In the embedding space, item similarity is encoded by clustering catalog SKUs into groups in the embedding space, by placing similarly related items close to each other in the embedding space. Catalog items are then searched for in the embedding, with the closest clusters searched for a particular catalog item.

Abstract: Certain aspects and features of this disclosure relate to automatic generation of composite images. For example, a method involves producing a representative image corresponding to a composite image based on a presentation context of input objects and segmenting the generated objects from the representative image to extract the generated objects from the representative image. The method also includes generating an inferred disposition of each of the generated objects and transforming each of the input objects to the inferred disposition of a corresponding generated object. The method can also include transmitting, storing, display, or rendering, in response to the transforming, the composite image of the input objects. Certain aspects and features also include computer systems, apparatus, and computer programs recorded on one or more computer storage devices, each configured to perform the actions of the method.

Abstract: Methods and systems are provided for using machine learning to efficiently promote eco-friendly products. In embodiments described herein, a product descriptions associated with a product is obtained. The product description includes subject matter indicating an environmental effect of the product. Thereafter, a score for the product correlated to the environmental effect of the product is generated by a machine learning model based on the product description of the product. The score is then provided for presentation to a user to indicate the correlated environmental effect of the product.

Abstract: A search system determines a hierarchy of attributes for a set of search results and returns the search results based on a top-ranked attribute. The search system identifies item listings based on a search input and determines attributes of the item listings. A machine learning model generates a hierarchy of the attributes. The search results are provided based on a top-ranked attribute from the hierarchy of the attributes.

Abstract: Some embodiments described herein relate to systems and methods for parameter-based synthetic model generation and recommendations including an image generation module and a recommendation module. The image generation module can receive one or more parameters and, responsive to receive the one or more parameters, generate a parameterized image using a generative machine learning model. The generative ML model may use the parameters as a seed for generating the parameterized image. The recommendation module may generate a first set of recommendations for a user of the client device and receive the one or more parameters. The recommendation module may determine, based on the one or more parameters, a second set of recommendations for the user of the client device. The second set of recommendations may include at least one element from the first set of recommendations.

Abstract: An object affinity determination and scoring system is described that is configured to support control by object providers in locating related objects. In a first example, an affinity system supports generation of affinity rules through interaction with a rule generation user interface. In a second example, the affinity system supports training and retraining of a machine-learning model to generate the affinity score. In a third example, the affinity scoring module supports output of a user interface having an input portion that supports user interaction to determine an affinity of selected objects to each other.

Abstract: The present disclosure relates to systems, non-transitory computer-readable media, and methods for generating and providing recommendations to view items in store by providing item categorization, physical store traffic modelling, historical analysis of returns, and inventory data to a delayed in-situ collaborative filter recommendation engine. In particular, in one or more embodiments, the disclosed systems receive selection of an item to purchase online and pick up in store from a client device. In response, in one or more embodiments, the disclosed systems determine item categorization, accesses physical store traffic modelling, and/or generates an analysis of historical return of items. Further, in one or more embodiments, the disclosed systems utilize a delayed in-situ collaborative filter recommendation engine to determine a recommendation of an additional item to view in store.

Abstract: Systems and methods for inferring compatibility relationships are described. Embodiments of the present disclosure identify user interaction history including an interaction between a user and a first product, wherein the first product comprises an attribute that is compatible with a subset of available products; query a database that includes the available products to identify a second product from the subset of available products based on the attribute, wherein the second product is identified based on a knowledge graph that includes a first node representing the first product and a second node representing the second product; and provide a customized user experience for the user that indicates the second product and the attribute.