An Effective Emotional Analysis Method of Consumer Comment Text Based on ALBERT-ATBiFRU-CNN

Abstract

To address the challenges of insufficient feature extraction for text sentiment analysis in the e-commerce big data environment, the author proposes a deep learning-based emotion analysis method of consumer comment text. Firstly, the author obtained the contextualized word vectors by using a pretrained language model called A Lite Bidirectional Encoder Representations From Transformers (ALBERT). Secondly, the researcher used the bidirectional gate recurrent unit (BiGRU) model to capture the semantic information through the combination of positive and negative directions, measure the emotional polarity information of each text as a whole, and then catch the local characteristic information of the text using the convolutional neural network (CNN) model. Finally, the author calculated the weight distribution through the attention mechanism. The experiments on a publicly available consumer review dataset showed that the recall, precision, and F1-score of the proposed text emotion analysis method were 0.9417, 0.9552, and 0.9484, respectively, which are higher than the existing methods. Therefore, the proposed method is of great significance in capturing the emotions of consumers on e-commerce platforms.