Meta Platforms, Inc.
Evaluating content items based upon semantic similarity of text

Abstract:

A content review system for an online system automatically determines if received content items to be displayed to users contain text that violates a policy of the online system. The content review system generates a semantic vector representing semantic features of text extracted from the content item, for example, using a neural network. By comparing the semantic vector for the extracted text with stored semantic vectors of extracted text previously determined to violate one or more policies, the content review system determines whether the content item contains text that also violates one or more policies. The content review system also reviews stored semantic vectors previously determined to be unsuitable, in order to remove false positives, as well as unsuitable semantic vectors that are sufficiently similar to known suitable semantic vectors and as such may cause content items having suitable text to be erroneously rejected.