Text Clustering in Social Media

Abstract

With the increasing popularity of social media platforms like Twitter which is an important data source. Analyzing user-generated content become an important research area in the field of Natural Language Processing (NLP) and Machine Learning (ML). One of NLP and ML applications is text based emotion detection (TBED), which is a challenging task due to the informal nature of the text and the limited context provided specially the dialectical one. The main contribution of this work lies in the development and evaluation of an ensemble clustering approach for automatic labeling of text data according to the Ekman emotional model (happy, sad, angry, disgust, fear and surprise) from text in Algerian dialect derived from Twitter. The proposed method combines multiple models of clustering algorithm to produce a single prediction. We utilized a pre-trained BERT (Bidirectional Encoder Representations from Transformers) model specifically designed for multilingual text (including Arabic dialects), for the representation of the text in a dense vector space. The combination of ensemble clustering techniques with a multilingual BERT model shows promise in accurately capturing the nuanced emotions expressed in Algerian tweets. The findings of this research have implications for understanding public emotions, improving customer satisfaction analysis, and enhancing social media monitoring tools.