Dual channel Chinese sentiment analysis of characters and words based on deep learning

Abstract

In recent years, sentiment analysis has been a hot research topic in the field of natural language processing. In this field, in view of the complex Chinese semantics in Chinese sentiment analysis tasks, the traditional sentiment analysis methods have insufficient effective information extraction and low classification accuracy, and there is room for further improvement in disambiguation. This paper proposes a sentiment analysis model (CWBCNN-Att) based on the fusion of Chinese character vectors and word vectors and an attention mechanism. Among them, the dimension of characters is added to the traditional word-based analysis method, which can extract richer local information, alleviate the problem of unregistered words in the vocabulary, and reduce the problem of ambiguity in Chinese texts to a certain extent; Secondly, this paper uses two parallel and independent bidirectional LSTM and GRU layers to extract the feature information of words respectively to speed up the processing; Furthermore, an attention mechanism is applied to the output of the bidirectional layers of our model to emphasize different weights for different words. To reduce the dimension of features and extract location-invariant local features, we utilize convolution and pooling mechanisms. The experimental results show that the method has excellent performance in terms of accuracy and recall.