Sentiment analysis of Chinese game reviews based on SKEP

Abstract

As China's game industry enters the era of the stock market, improving the quality of game products in all aspects is gradually becoming an important part of the development of China's game industry. There exists a large amount of game review data on the Internet, and sentiment analysis of them can provide important help for game makers to make decisions. However, the huge scale of game reviews and the extensive use of metaphors and sarcasm make it difficult to conduct sentiment analysis. In this paper, we experiment with a publicly available Chinese game review dataset using the ERNIE pre-training model equipped with the SKEP method, which improves the model's masking strategies and pre-training objectives to specialize the model's sentiment analysis ability. In addition, we compared the performance of ERNIE1.0, RoBERTa, MacBERT, BERT-wwm, and Nezha. The precision, recall, and F1 score of SKEP are higher than those of the models for comparison studies, which confirms the effectiveness of SKEP in this area.