Depressive Bangla Text Detection from Social Media Post Using Different Data Mining Techniques

Abstract

Nowadays, social media have opened a door in front of us to share individual’s expressions, emotions, and attitudes toward any incident. Sentiment analysis from different social media posts may help us to detect positive, negative, or emotional behavior toward society. Depressive text detection from social media posts is one of the most challenging parts of individual behavior or sentiment analysis. In this paper, different machine learning algorithms are used to detect depressive Bangla text from social media posts. Pre-processing steps like stemming, stop word removal, etc., are used to clean the collected data, and feature extraction techniques like count vectorization, TF-IDF, word embedding, etc., are applied to the collected dataset which consists of 6178 texts collected from social media posts. We have achieved the highest 97% classification accuracy using decision tree and 94% accuracy for bidirectional LSTM (deep learning model) to predict depressive text in Bangla language. Depressive text detection from social media posts will create an opportunity for psychologists to analyze sentiment from shared posts, reactions, and attitudes which may lessen the unwanted activities of the depressed people through diagnosis and taking them under treatment.