Extraction of Relations Between Entities from Human-Generated Content on Social Networks

In this work, we present a method to extract new knowledge from content shared by users on social networks, with particular emphasis on extraction of evolving relations between entities. Our method combines natural language processing and machine learning for extracting relations in the form of triples (subject-relation-object). The method works on domain-specific content shared on social networks: users can define a domain through a set of criteria (social networks accounts, key-words or hashtags) and they can define a limited set of relations that are of interest for the given domain. Based on this input, our method extracts the relevant triples for the domain. The method is demonstrated on content retrieved from Twitter, belonging to different domain-specific scenarios, like fashion and chess. Results are promising, in terms of both precision and recall.