A New Approach to Testing BGRU Text Affective Analysis
Download as PDF
DOI: 10.25236/AISCT.2019.084
Corresponding Author
Tian Li
Abstract
As an indispensable part of social life, affective analysis of text data generated by social networks has become a research hotspot in the field of natural language processing. In view of the fact that depth learning technology can automatically construct text features, CNN (Conventional Neural Network) and BLSTM (Bidirectional Long Short-term Memory) have been proposed to solve the problem of text emotion analysis. However, BGRU (Bidirectional Gated Recurrent Unit) can memorize the context information of the sequence, and has simple structure and fast training speed. This thesis proposes a BGRU-based affective analysis method for English texts. Firstly, the text is converted into a sequence of word embeddings, then the contextual affective features of the text are obtained by using BGRU, and finally the affective tendencies of the text are given by the classifier. Experiments on ChnSentiCorp corpus show that this method achieves 90.61% F1 value, which is superior to CNN and BLSTM models, and the training speed is 1.36 times as fast as BLSTM.
Keywords
Bidirectional Gated Recurrent Unit (BGRU); Deep learning; Affective analysis