A multi-task learning approach for consumer sentiment classification and emotion detection via behavioral data

Sentiment analysis captures overall opinion polarity, but it often fails to reveal the key emotions that drive consumer behavior. The recognition of both sentiment and emotions provides crucial information to uncover underlying drivers of consumers behavior and brand loyalty. These important insights enable businesses to detect emerging issues, refine products and services, and personalize consumer experiences. This study proposes a multi-task approach that performs sentiment classification and emotion detection tasks based on consumer reviews and corresponding ratings. To this end, we propose Consumer Sentiment and Emotion Modeling ( COSMO-MT ), a deep learning-based multi-task model for sentiment classification and emotion detection. It begins with an input layer followed by embedding layers that provide sentiment-specific, emotion-specific, and shared embeddings from the input text. Each embedding is processed through shared layers consisting of CNN, BiLSTM, and attention mechanisms. Moreover, several auxiliary features are incorporated into the shared layers to provide explicit linguistic information. The outputs of the shared layers are then passed to task-specific layers for sentiment classification and emotion detection. Finally, corresponding dense and output layers generate predictions for both tasks. We evaluate the efficacy of the proposed COSMO-MT model on two publicly available datasets. It significantly outperforms existing studies and baseline methods.