Entity extraction for people profile matching

Abstract

With the rise in social media platform usage, the average number of people profiles has increased as well. The fact that people have social media profiles on multiple platforms reveals the interesting problem of matching them in order to aggregate all the profiles in one. As a significant part of these profiles are made of unstructured text, extracting labeled data using an entity extraction system can lead to an increase in precision and recall. This thesis documents the effects a bootstrapped pattern learning approach can have on a profile matching system. The entity extraction system is trained in a distributed manner on a big amount of data, in order to generate as many quality patterns as possible.