A Generic Model Based on Multiple Domains for Sentiment Classification

Traditional models for sentiment classification are trained and tested on the same dataset. However, the model parameters trained on one dataset are not suitable for another dataset and it takes much time to train a new model. In this paper, we propose a generic model based on multiple domains for sentiment classification (DCSen). In DCSen, domain classification is used to generalize the sentiment classification model, so the trained model’s parameters can be applied to different datasets in given domains. Specifically, the document is first mapped to the domain distribution which is used as a bridge between domain classification and sentiment classification, and then sentiment classification is completed. In order to make DCSen more generic, the sentiment lexicon is introduced to select the sentences in a document and the more representative datasets are obtained. For the purpose of improving accuracy and reducing training time, transfer learning based on neutral networks is used to get the document embeddings. Extensive experiments on the datasets of 15 different domains show that DCSen can achieve better performance compared with traditional models in the aspect of generality.