Abstract:Sentiment analysis is an essential part of the natural language processing domain. Currently, sentiment analysis is mainly applied to scenarios such as user tweets, film comments, and opinion analysis. These scenarios generally contain sufficient training data. However, for some scenarios that lack sufficient training data, their performance is often limited. For this reason, a sentiment analysis method based on transfer learning was proposed in this study. With self-attention and adversarial learning as the core framework, the method could be used to assist the learning of small-scale specific domain sentiment analysis by using sentiment data from the rest of the large-scale domains, thus achieving better sentiment analysis results in small-scale specific domains. In the experimental part, the electric power field was taken as an example. The relevant experimental results demonstrate the effectiveness of the proposed method.

Sentiment Analysis of Electric Power Domain based on Transfer Learning