Extracting social values and group identities from social media text data

Abstract

This paper presents preliminary results on the extraction of group identities from social media data using topic models and a rich form of sentiment analysis that is designed to correspond to psychologically-validated emotional states. Our approach is based upon the sociological notion that group identity forms the basis for behavioral change [1]. We begin by inferring social values from social media text data by combining information regarding topic content and sentiment. Next, groups are inferred as a latent variable mediating between individual social media authors and social values. A topic model is proposed, extending the Ailment Topic Aspect Model (ATAM) used by Paul and Dredze [2], and applied to a large set of blog data extracted from the Media Cloud [3] daily updates. We also provide a qualitative and quantitative analysis of model outputs.

Publication
Multimedia Signal Processing (MMSP), 2012 IEEE 14th International Workshop on