Research on Sentiment Analysis Fusion Based on Neural Network
Yuanyuan Ma, Yongyong Sun, Fei Xu · 2023
With the aim of overcoming the limitations in feature extraction capability, OOV (out-of-vacabulary) and polysemous word in Chinese text sentiment analysis, dual attention gated recurrent neural network (AW-Att-BiGRU) with word and character information fusion was proposed. Firstly, character and word level vector are taken as input to the model to obtain semantic representation to the maximum extent. Then, characters and words with obvious emotional characteristics are given higher weights through the attention mechanism, to achieve a more comprehensive understanding of the emotional tendencies within the text and enhance the accuracy of sentiment analysis. Based on the experimental results using the Tan Songbo Hotel review dataset, the proposed model achieved an accuracy rate of 94.56%, and the accuracy rate on Weibo review dataset is 93.87%, which improves the performance compared with a single model and has certain application value.