Enhancing Sentiment Analysis Using Attention Mechanism

Abstract

The rise of social platforms in China has increased user-generated content (UGC), which often contains rich emotional information. Analyzing UGC sentiment helps identify group emotions on specific events, aiding decision-making for platforms and authorities. This paper introduces a novel sentiment analysis method that combines a bipolar word attention mechanism with a DSA-CNN model. By developing bipolar word vectors and leveraging an emotional dictionary, the model extracts features from positive and negative emotions, enhancing short text emotion detection. Experiments on three datasets demonstrate superior performance and faster convergence due to the attention mechanism. Future work aims to improve the sentiment dictionary and word vector training to address current limitations and better capture contextual influences on sentence emotion.