Abusive Text Examination Using Latent Dirichlet Allocation, Self Organizing Maps and K Means Clustering

The widespread misuse of social media to disseminate hate speech targeted at a particular individual, community, religion, race, sex or caste has engendered researchers from across the world to formulate strategies and methodologies to counter this menace. Without detection and analysis of hate speech, one cannot imagine the social media to be free of malicious content. This paper proposes a methodology which employs a combination of popular topic modeling technique i.e. Latent Dirichlet Allocation (LDA) and an unsupervised machine learning technique i.e. self-organizing maps (SOM) to analyze hate speech spread over social media. This method is compared to K means clustering used after the application of LDA. Both the techniques used together provide a powerful analysis. The proposed LDA model outputted ten topics for features and had a low perplexity with a higher negative log-likelihood score.