A Neural Network Based on WXLNet and Multi-Task Lable Embedding for Sentiment Analysis

Sentiment analysis is an important field in natural language processing, the development of which has been greatly promoted by deep learning recently. However, previous studies focused on the structure of the model, and did not make full use of text semantics and label information. Moreover, the traditional model does not perform well in complex fine-grained sentiment classification. In this paper, we propose a new deep neural network model WXLNet-MTLE. By modifying XLNet language model, the utilization of text information and the language expression ability have been greatly improved. At the same time, we add Multi-Task Lable Embedding to improve the generalization ability of our model in its downstream sentiment analysis tasks. Comparative analysis are carried out on the data sets of 3 different tasks and 8 different scenarios. The experimental results show that WXLNet-MTLE performs better in sentiment analysis of multiple scenes than the other models.