Based BERT-BiLSTM-ATT Model of Commodity Commentary on The Emotional Tendency Analysis

Abstract

In order to realize the analysis of the emotional tendency of the product user reviews. This paper proposes a method for analyzing the sentiment orientation of product reviews based on the BERT-BiLSTM-ATT model. First, use the BERT model to obtain the feature representation of the product review text, and then input the obtained feature representation into the BiLSTM network to extract the emotional features of the product review. Add an Attention layer before the output layer of the normal BiLSTM model to further improve the classification accuracy, and finally combine with Softmax The classifier classifies the extracted features. To validate the algorithm, the design and LSTM, BiLSTM, BiLSTM-ATT, BERTBiLSTM comparative model experiments, experimental results show that the algorithm accuracy on test set are improved 6.17%, 3.75%, 2.83%, 1.03%, to prove the effectiveness of this method in the relevant classification task.