On the Detection of Hate Speech, Hate Speakers and Polarized Groups in Online Social Media

Abstract

The objective of this dissertation is to explore the use of machine learning algorithms in understanding and detecting hate speech, hate speakers and polarized groups in online social media. Beginning with a unique typology for detecting abusive language, we outline the distinctions and similarities of different abusive language subtasks (offensive language, hate speech, cyberbullying and trolling) and how we might benefit from the progress made in each area. Specifically, we suggest that each subtask can be categorized based on whether or not the abusive language being studied 1) is directed at a specific individual, or targets a generalized ``Other" and 2) the extent to which the language is explicit versus implicit. We then use knowledge gained from this typology to tackle the ``problem of offensive language" in hate speech detection. A key challenge for automated hate speech detection on social media is the separation of hate speech from other instances of offensive language. We present a Logistic Regression classifier, trained on human annotated Twitter data, that makes use of a uniquely derived lexicon of hate terms along with features that have proven successful in the detection of offensive language, hate speech and cyberbullying. Using the tweets classified by the aforementioned hate speech classifier, we extract a set of users for which we collect demographic and psychological attributes, with the goal of understanding how these attributes are related to hate speech use. We first present a binary Random Forest classifier for predicting whether or not a Twitter user is a hate speaker. We then explore the use of linear and Random Forest regression models as a means of explaining and predicting levels of hate speech use based on user attributes. To the best of my knowledge, this work is the first to present an automated approach for detecting individual hate speakers. Finally, we present a non-negative matrix factorization (NMF) algorithm for identifying polarized groups using tripartite graphs (user-post-tag) gleaned from social media data. This work is heavily inspired by the need for an unsupervised approach that works well in contexts varying in the nature of the controversy, the level of polarization, the number of polarity groups involved, and the presence of neutral entities. I present the first ever analysis of polarization on data from the Tumblr platform, showing improved performance over traditional community detection methods and the state-of-the-art method of NMF on bipartite graphs.