A Commodity Review Sentiment Analysis Based on BERT-CNN Model

Sentiment analysis through the investigation on commodity reviews will be of great importance to commodity quality improvement of the seller and subsequent consumption choice of buyers. The accuracy of the existing sentiment analysis models remains to be further improved, so a BERT-CNN sentiment analysis model, an improvement of the original BERT model, was proposed in this paper in order to improve the accuracy of commodity sentiment analysis. Firstly, BERT model was constructed, and then a representation layer was input into the model to encode the review texts; after then, CNN semantic extraction layer was utilized to extract local features of the review text vectors, BERT semantic extraction layer to extract global features of the review text vectors and semantic connection layer to fuse features extracted by the two complementary models; in the end, a sentiment analysis of online commodity reviews was performed via the sentiment classification layer. The experimental results indicated that in comparison with BERT and CNN models, F1 value of BERT-CNN model was elevated by about 14.4% and 17.4%, respectively.