Speaker: Dr. Einat Minkov
Title: SocialVec: Social entity embeddings
Zoom Link
Abstract:
We introduce a general framework for eliciting social world knowledge from social networks, and demonstrate its application to the social network of Twitter. Assuming that popular Twitter accounts represent entities of general interest, SocialVec learns low-dimensional embeddings of these accounts from their co-occurrence patterns in social media. Similar to word embeddings, which facilitate tasks that involve text processing, we expect social entity embeddings to benefit information processing tasks of social flavor. We learn social entity embeddings for roughly 200,000 popular entities in Twitter from a sample of the Twitter network, which includes more than 1.3 million users, and the accounts that they follow, and evaluate the resulting embeddings in two different case studies. The first study applies to the task of automatically inferring the personal traits of users from their social media profiles. In another case study, we exploit SocialVec embeddings for gauging the political bias of news sources in Twitter. In both cases, we demonstrate high performance using SocialVec embeddings, and prove it advantageous compared with other popular entity embedding schemes. We make the SocialVec entity embeddings publicly available to support further exploration of social world knowledge as reflected in Twitter. This is joint work with Nir Lotan.
Bio:
Einat is a Senior Lecturer at the Information Systems Dep., University of Haifa. She received a BSc and MSc in Industrial Engineering in 2000 from Tel-Aviv University, and a PhD in Language Technologies from Carnegie Mellon University in 2009. During and after her PhD, she worked in the research labs of Microsoft, Redmond WA and Nokia, Cambridge MA, USA. In 2019-20, she visited the research lab of Google in Tel-Aviv. Her research interests include natural language processing, semantics and machine learning. She serves in the reviewing boards of leading conferences and journals in these areas, such as ACL, EMNLP, WWW, IJCAI, ICML, NuerIPS and the Transactions of the Association for Computational Linguistics (TACL).
