An External Knowledge-Based Sentiment Classification Method for Chinese Short Texts

With the rapid development of Internet applications, a large number of short texts expressing public opinion with emotional views have been generated. The subtexts of these short texts are complex, which bring a great challenge to emotional classification. In order to solve this problem, this paper proposes a method for sentiment classification of Chinese short texts based on fused external knowledge. Firstly, we construct a prompt text with fused sentiment knowledge, splice the prompt text with the original input text into the pre-training language model, and guide the pre-training model through the prompt. Then, we design label words for the text domain and map the prediction results for classification. The experimental results show that this method exhibits good performance on the Meituan Waimai dataset, with an accuracy of 81.3%, which are 1.8%, 2.2% and 3.5% higher than the mainstream modeling methods such as BERT, TextCNN and ERINE, respectively.