Graph based tweet entity linking using DBpedia

Abstract

Twitter has became an invaluable source of information, due to his dynamic nature with more than 400 million tweets posted per day. Determining what an individual post is about can be a non trivial task because his high contextualization and his informal nature. Named Entity Linking (NEL) is a subtask of information extraction that aims to ground entity mentions to their corresponding node in a Knowledge Base (KB), which requires a disambiguation step, because many resources can be matched to the same entity that lead to synonymy and polysemy problems. To overcome these problems, especially in the context of short text, we present a novel system for tweet entity linking based on graph centrality and DBpedia as knowledge base. Our approach relies on the assumption that related entities tend to appear in the same tweet as tweets are topic specific. Also, we address the problem of irregular name mentions. Finally, to show the effectiveness of our system we evaluate it using a real twitter dataset and compare it to a well known state-of-the-art named entity linking system for short text.