Pinterest, Inc.

Disclosed are systems and methods that generate a natural language prompt that is configured to be processed by a generative model, such as a large language model (LLM), and includes certain user information to facilitate the determination and/or generation of customized content for users of an online platform. For example, textual information associated with certain user information may be extracted and aggregated and incorporated into one or more natural language prompts, which may be processed by a generative model, such as an LLM, to generate a particular output based on the type of customized content being sought and/or generated for the user. The output may then be processed to determine and/or generate the customized content or the user.

Systems and methods for generating user notifications to a set of users of a social networking service is presented. For each user of a set of users of the social networking service, one or more machine learning models selects an optimal notification channel, an optimal notification template, and optimal personalization content for configurable elements of a selected notification template. Each of these determinations/selections is made according to and based on a likelihood of increased user engagement with the social networking service. Upon determining the notification channel, notification template, and personalizations to the template, the notification is generated and sent to the corresponding user.

Described are systems and methods to determine a content item recommendation that is included in a content item bid request, recommending to a third party, a third party content item to include in a content item bid that is responsive to the bid request. The content item recommendation may indicate a particular third party product or third party content item that is predicted to perform well in a content item slot and for a specific user without disclosing user information to the third party.

Systems and methods for identifying relevant content within a corpus of visual content items in response to a user's text-based query are presented. In response to a text-based query, the query is mapped to a most-engaged content item of the corpus of visual content items included in responses to the query from a plurality of users. At least one text-based term associated with the most-engaged content item is identified and combined with the query from an expanded query. The expanded query is mapped to an interest node of an interest taxonomy and content items associated with the mapped interest node are identified. At least some of the content items associated with the mapped interest node are selected and returned as response content to the received query.

Systems and methods for generating user notifications to a set of users of a social networking service is presented. For each user of a set of users of the social networking service, one or more machine learning models selects an optimal notification channel, an optimal notification template, and optimal personalization content for configurable elements of a selected notification template. Each of these determinations/selections is made according to and based on a likelihood of increased user engagement with the social networking service. Upon determining the notification channel, notification template, and personalizations to the template, the notification is generated and sent to the corresponding user.

Systems and methods for identifying relevant content within a corpus of visual content items in response to a user's text-based query are presented. In response to a text-based query, the query is mapped to a most-engaged content item of the corpus of visual content items included in responses to the query from a plurality of users. At least one text-based term associated with the most-engaged content item is identified and combined with the query from an expanded query. The expanded query is mapped to an interest node of an interest taxonomy and content items associated with the mapped interest node are identified. At least some of the content items associated with the mapped interest node are selected and returned as response content to the received query.

Systems and methods for identifying content for an input query are presented. A mapping model is trained to map elements of an input query embedding vector for a received query into one or more elements of a destination embedding vector. In response to receiving an input query, an input query embedding vector is generated that projects into an input query embedding space. The input query embedding vector is processed by the mapping model to map the input query embedding vector into one or more elements of a destination embedding vector in a destination embedding space, resulting in a partial destination embedding vector. Items of a corpus of content are projected into the destination embedding space and the partial destination embedding vector is also projected into the destination embedding space. A similarity measure determines the most-similar items to the partial destination embedding vector and at least some of the most-similar items are returned in response to the input query.

Described is a system and method for enabling dynamic selection of a search input. For example, rather than having a static search input box, the search input may be dynamically positioned such that it encompasses a portion of displayed information. For example, a user may touch a touch-based display using two fingers to invoke the dynamic search input and then determine a size and a position of the dynamic search input by moving their fingers on the display. An image segment that includes a representation of the encompassed portion of the displayed information is generated and processed to determine an object represented in the portion of the displayed information. Additional images with visually similar representations of objects are then determined and presented to the user.

Systems and methods for responding to a subscriber's text-based request for content items are presented. In response to a request from a subscriber, word pieces are generated from the text-based terms of the request. A request embedding vector of the word pieces is obtained from a trained machine learning model. Using the request embedding vector, a set of content items, from a corpus of content items, is identified. At least some content items of the set of content items are returned to the subscriber in response to the text-based request for content items.

Described are systems and methods that resolve the shortcomings of existing recommendation systems that focus on immediate user reward rather than long-term user satisfaction. The disclosed implementations increase user session duration and user session depth during a session by selecting items that incent the user to remain engaged and participating in the session, rather than optimizing for immediate user metrics.