Psychology-informed Natural Language Understanding: Integrating Personality and Emotion-aware Features for Comprehensive Sentiment Analysis and Depression Detection

This paper presents a novel approach to natural language understanding, integrating personality and emotion-aware features for sentiment analysis and depression detection. This research aims to enhance the performance of natural language understanding tasks, specifically sentiment analysis and depression detection, while also promoting explainability by including interpretable insights into psychological factors, such as emotion and personality, that influence these tasks. We refer to this additional feature as the psychology-informed module, alongside attention and transformer models. We achieved a significant improvement in accuracy using only the emotion feature: 3.4% for sentiment analysis on the IMDb dataset and 3.1% for depression detection on the SDCNL dataset. Similarly, using the personality feature led to a 2.5% improvement in sentiment analysis on the Polarity dataset and a 2.9% improvement in depression detection on the SDCNL dataset. On the other hand, the culmination of combining both psychological features achieves an accuracy of 0.8775 and 0.9053 for sentiment analysis on the Polarity and IMDb datasets, respectively. Additionally, notable results were obtained for depression detection, with accuracies of 0.8533 and 0.7177 on the Twitter (now known as X platform) and SDCNL datasets, respectively. These advancements enhance model accuracy and improve explainability, fostering versatile real-world applications. We thoroughly examined the factors, advantages, and limitations associated with this approach (psychology-informed module), providing a comprehensive discussion within the scope of our study. The findings pave the way for future research to explore innovative techniques, further expanding the interdisciplinary impact of psychology-informed natural language understanding.