Abstract: Systems and method for determining a topic cohesion measurement between a content item and a hyperlinked landing page are presented. In one embodiment, a plurality of content item signals is generated for the content item and a corresponding plurality of signals are generated for the hyperlinked landing page. An analysis of the corresponding signals is conducted to determine a measurement of topic cohesion, a topic cohesion score, between the content item and the hyperlinked landing page. A cohesion predictor model is trained to generate the predictive topic cohesion score between an input content item and a hyperlinked landing page. Upon a determination that the topic cohesion score is less than a predetermined threshold, remedial actions are taken regarding the hyperlink of the content item. Alternatively, positive actions may be carried out, including promoting the content item to others, associating advertisements with the content item, and the like.

Abstract: Methods, systems, and apparatus, including computer programs encoded on computer storage media for receiving a set of images at a social media system, wherein each image includes one or more recognized features associated with one or more lightness values; indexing each image using the one or more recognized features and the associated range of lightness values; receiving a query; determining a first group of images that is responsive to the query; determining that the query triggers a lightness filter to be displayed on the user device; providing the first group of images for display on a user interface with one or more lightness filter indicators; and in response to a user selection of one of the one or more lightness filter indicators: filtering the first group of images to determine a filtered group of images, and updating the images provided for display according to the filtered group of images.

Abstract: Described are systems and methods to determine hair patterns presented in content items. The determined hair patterns may be associated with the content items to facilitate indexing, filtering, etc. of the content items based on the determined hair patterns. In exemplary implementations, a corpus of content items may be associated with an embedding vector that includes a binary representation of the content item. The embedding vectors associated with each content item can be provided as inputs to a trained machine learning model, which can process the embedding vectors to determine one or more hair patterns presented in each content item while eliminating the need for performing image pre-processing prior to determination of the hair pattern(s) presented in the content item.

Abstract: Disclosed are implementations that enable the linking or connection of objects and different scenes in which those objects are represented. For example, a corpus of scenes (e.g., digital images) that include a representation of one or more objects may be processed using the disclosed implementations to segment from those scenes the individual objects represented in those scenes. The disclosed implementations may further determine clusters of visually similar object segments and form object clusters for those object segments. The scenes that include those object segments are also linked to the object cluster. With scenes linked to different object clusters, a user may select one or more query objects or a query scene and be presented with other scenes that include visually similar objects, even though the overall scenes may be visually different.

Abstract: Systems and methods are presented for inferring an embedding vector of an item of a first type into the embedding space. Upon receiving a first time for which there is no embedding vector, documents of a document corpus that include (co-occurrence) both the received item and other items of the same type are identified. Of those other items that have embedding vectors, those embedding vectors are retrieved and averaged. The resulting averaged embedding vector is established as an inferred embedding vector for the received item.

Abstract: Systems and methods for generating embeddings for nodes of a corpus graph are presented. The embeddings correspond to aggregated embedding vectors for nodes of the corpus graph. Without processing the entire corpus graph to generate all aggregated embedding vectors, a relevant neighborhood of nodes within the corpus graph are identified for a target node of the corpus graph. Based on embedding information of the target node's immediate neighbors, and also upon neighborhood embedding information from the target node's relevant neighborhood, an aggregated embedding vector can be generated for the target node that comprises both an embedding vector portion corresponding to the target node, as well as a neighborhood embedding vector portion, corresponding to embedding information of the relevant neighborhood of the target node. Utilizing both portions of the aggregated embedding vector leads to improved content recommendation to a user in response to a query.

Abstract: Methods, systems, and apparatus, including computer programs encoded on a computer storage medium, for providing content. One of the methods includes providing a plurality of image content items to an application interface of a client device for presentation; receiving a user selection of a particular image content item of the plurality of presented image content items; and responsive to the selection, providing a combination of native content and third party content associated with the selected image content item, wherein the native content includes a close up view of the selected image content item and the third party content includes a third party webpage.

Abstract: Described is a system and method for enabling visual search for information. With each selection of a search term, additional search terms are dynamically selected and presented to the user in conjunction with results matching the currently selected search terms. Likewise, a selected search term may be tokenized and a graphical token presented to the user to represent the selected search term.

Abstract: Systems and methods for efficiently training a machine learning model are presented. More particularly, using information regarding the relevant neighborhoods of target nodes within a body of training data, the training data can be organized such that the initial state of the training data is relatively easy for a machine learning model to differentiate. Once trained on the initial training data, the training data is then updated such that differentiating between a matching and a non-matching node is more difficult. Indeed, by iteratively updating the difficulty of the training data and then training the machine learning model on the updated training data, the speed that the machine learning model reaches a desired level of accuracy is significantly improved, resulting in reduced time and effort in training the machine learning model.

Abstract: Disclosed are systems and methods for autonomously extracting attributes from domains of a vertical. The disclosed implementations train a deep neural network (“DNN”) based on one or more domains of a vertical using labeled embedding vectors generated for nodes of those one or more domains. The trained DNN may then be used to autonomously label nodes of other domains within the same vertical such that attributes corresponding to those labels can be extracted.

Abstract: This disclosure describes systems and methods for establishing promotions for sellers and promoting images of items to users on behalf of sellers. A management service receives a source location identifier from a seller, processes images stored in an image data store to determine images that include the source location identifier in the corresponding image information and includes those images in a promotion that is established for the seller. Likewise, the management service may determine users that have previously interacted with the images and include those users in the promotion campaign.

Abstract: This disclosure describes systems and methods that facilitate the generation of recommendations by traversing a graph. Walks that traverse the graph may be initiated from a plurality of different nodes in the node graph. In order to give greater or lesser weight to particular nodes, the walks may have different lengths depending on the nodes from which they are initiated, or an unequal amount of walks may be distributed between nodes from which walks are initiated. A plurality of walks through a node graph may be tracked, and visit counts or scores for nodes in the node graph may be determined. For example, scores may be increased for nodes that are visited by a walk initiated from a first node and a second walk initiated from a second node, or scores may be decreased for nodes that are not visited by a first walk initiated from a first node and a second walk initiated from a second node. Content corresponding to nodes may be recommended based on the scores or visit counts.

Abstract: Methods, systems and apparatus, including computer programs encoded on computer storage media for receiving a set of images at a social media system, wherein each image includes one or more recognized features associated with one or more lightness values; indexing each image using the one or more recognized features and the associated range of lightness values; receiving a query; determining a first group of images that is responsive to the query; determining that the query triggers a lightness filter to be displayed on the user device; providing the first group of images for display on a user interface with one or more lightness filter indicators; and in response to a user selection of one of the one or more lightness filter indicators: filtering the first group of images to determine a filtered group of images, and updating the images provided for display according to the filtered group of images.

Abstract: A system for and a method of using user-entered information to return more meaningful information in response to Internet search queries are disclosed. A method in accordance with the disclosed subject matter comprises managing a database in response to multiple user inputs and displaying search results from the database in response to a search query. The search results include a results list and supplemental data related to the search query. Managing the database includes, among other things, re-ranking elements in the results list, storing information related to relevancies of elements in the results list, blocking a link in the results list, storing links to documents related to the search query, or any combination of these. The supplemental data include descriptions of or indices to one or more concepts related to the search query.

Abstract: The described implementations are operable to determine potential objects of interest to a user based on a blend of the user's long-term behavior and short-term interests. Long term user behavior may be determined for the user over a period of time and represented as continuous data. Short-term interest may be determined based on objects with which the user has recently interacted and attributes of those objects may be represented together as continuous data corresponding to the short-term user interest. The continuous data of the short-term interest and long-term user behavior may be blended to produce a user embedding. The user embedding may then be compared with objects to determine objects that are of potential interest to the user.

Abstract: Systems and methods for recommending content to an online service user are presented. In response to a request from a user, a set of n-grams of the request are generated, with each n-gram comprising one or more terms from the request and each n-gram of the set of n-grams being unique. Embedding vectors projecting the n-grams into a content item embedding space are generated, and the embedding vectors are combined into a representative embedding vector for the request. The nearest content items are identified according to a distance measure between a projection of the representative embedding vector and embedding vectors of content items of a corpus of content items in the content item embedding space. At least some of the nearest content items are returned as recommended content in response to the request from the user.

Abstract: Systems and methods are presented for inferring an embedding vector of an item of a first type into the embedding space. Upon receiving a first time for which there is no embedding vector, documents of a document corpus that include (co-occurrence) both the received item and other items of the same type are identified. Of those other items that have embedding vectors, those embedding vectors are retrieved and averaged. The resulting averaged embedding vector is established as an inferred embedding vector for the received item.

Abstract: Disclosed are implementations that enable the linking or connection of objects and different scenes in which those objects are represented. For example, a corpus of scenes (e.g., digital images) that include a representation of one or more objects may be processed using the disclosed implementations to segment from those scenes the individual objects represented in those scenes. The disclosed implementations may further determine clusters of visually similar object segments and form object clusters for those object segments. The scenes that include those object segments are also linked to the object cluster. With scenes linked to different object clusters, a user may select one or more query objects or a query scene and be presented with other scenes that include visually similar objects, even though the overall scenes may be visually different.

Abstract: Described are systems and methods for providing a multi-tasked trained machine learning model that may be configured to generate product embeddings from multiple types of product information. The exemplary product embeddings may be generated for a corpus of products (e.g., products included in a product catalog, etc.) based on both image information and text information associated with each respective product. Accordingly, the generated product embeddings may be compatible with learned representations of the different types of product information (e.g., image information, text information, etc.) and may be used to create a product index, which can be used to determine and serve product recommendations in connection with multiple different recommendation services that may be configured to receive different types of inputs (e.g., a single image, multiple images, text-based information, etc.).

Abstract: Described are systems and methods for providing a sequential trained machine learning model that may be configured to generate a user embedding that is representative of the user and is configured to predict a plurality of the user's actions over a period of time. The exemplary sequential trained machine learning model may be employed, for example, in connection with recommendation, search, and/or other services. Exemplary embodiments of the present disclosure may also employ the user embeddings generated by the exemplary sequential trained machine learning model in connection with one or more conditional retrieval systems that may include an end-to-end learned model, which are configured to generate updated user embeddings based on the user embeddings generated by the exemplary sequential trained machine learning model and certain contextual information.