Meta Platforms, Inc.
Search ranking and recommendations for online social networks based on reconstructed embeddings

Abstract:

In one embodiment, a method includes receiving a query inputted by a user; identifying one or more objects matching the query; generating a reconstructed embedding of the query based on one or more term embeddings associated with one or more of the n-grams of the query, respectively; calculating, for each identified object, a relevance-score based on a similarity metric of the reconstructed embedding of the query and an object embedding associated with the identified object; generating one or more search results based on the calculated relevance-scores, each search result corresponding to one of the identified objects; and sending a search-results interface for display, the search-results interface comprising one or more of the search results, each search result comprising a reference to its corresponding identified object.