Hybrid Deep Learning Approach for Sentiment Classification of Malayalam Tweets

Social media content in regional languages is expanding from day to day. People use different social media platforms to express their suggestions and thoughts in their native languages. Sentiment Analysis (SA) is the known procedure for identifying the hidden sentiment present in the sentences for categorizing it as positive, negative, or neutral. The SA of Indian languages is challenging due to the unavailability of benchmark datasets and lexical resources. The analysis has been done using lexicon, Machine Learning (ML), and Deep Learning (DL) techniques. In this work, the baseline models and hybrid models of Deep Neural Network (DNN) architecture have been used for the classification of Malayalam tweets as positive, negative and neutral. Since, sentiment-tagged dataset for Malayalam is not readily available, the analysis has been done on the manually created dataset and translated Kaggle dataset. The hybrid models used in this study combine Convolutional Neural Networks (CNN) with variants of Recurrent Neural Networks (RNN). The RNN models are Long Short-Term Memory (LSTM), Bidirectional LSTM (Bi-LSTM) and Gated Recurrent Unit (GRU). All these hybrid models improve the performance of Sentiment Classification (SC) compared to baseline models LSTM, Bi-LSTM and GRU.