Analysis of Word Embedding and Fusion Interactive Emotion

Text sentiment analysis is a significant research area within the field of natural language processing. The goal of sentiment analysis is to determine the sentiment polarity associated with each specific aspect or target in a given sentence. To enhance the richness of feature extraction and improve the model's performance in sentiment polarity prediction, this paper proposes a Word Embedding Fusion Interactive Sentiment Analysis Model. The model integrates Relational Word Embedding (RWE) and FastText embeddings, leveraging the complementary strengths of different word embeddings at the input layer to enrich feature representation. RWE encodes relational knowledge within word embeddings, while FastText captures subword-level features from text sequences. Additionally, the GloVe embedding is employed alongside an attention mechanism to learn the complementarity and enhancement relationships between the embeddings, generating a weighted distribution that enables adaptive learning of task-specific hidden representations. Finally, the model performs interactive learning between aspect terms and contextual representations to mine their mutual influence and fully utilize the semantic information between them, thereby improving the accuracy of sentiment polarity prediction.