Research And Application Of Product Fine-grained Opinion Mining Based On Deep Learning

With the rapid development of information technology,user-generated content has experienced explosive growth on the Internet.For e-commerce platform,massive user-generated content is the evaluation information of consumers on products,which contains consumers' opinions and emotional attitudes towards products.Use natural language processing technology can effectively mine fine-grained opinion information in product reviews,which will help merchants understand users and improve their own products as well as helping consumers understand products and chooseing their favorite product for themselves.Based on the method of deep learning,this paper focuses on the two tasks of extracting evaluation elements and fine-grained sentiment polarity analysis in fine-grained opinion mining for product evaluation to mine fine-grained opinion information of products.In terms of the extraction of evaluation elements,this paper builds a word pair relevance model based on the pre-training model and combines with the evaluation element bidirectional propagation iterative recognition framework A bidirectional iterative recognition algorithm for evaluation elements based on BERT,targeted at the problem of lacking considering of the collocation relationship between evaluation objects and evaluation words at the deep semantic level.In the experiment of extracting the evaluation elements of cosmetics and laptop products,this method is superior to the previous two-way propagation iterative recognition algorithm based on dependency syntax;in the recall strategy of low-frequency words in evaluation elements,it is proposed to use word mixing Vector semantic similarity recall method,this method considers the word segmentation error factor brought by the word segmentation tool,and improves the accuracy of the evaluation elements of the recall.In terms of fine-grained sentiment polarity analysis,this paper designs a fine-grained sentiment polarity classification model based on text matching for the problem of insufficient semantic interaction between the target evaluation dimension text and the product evaluation text.The model is designed from the perspective of text matching,which strengthens the interaction process between target evaluation dimension text and product evaluation text,and improves the accuracy of fine-grained emotion polarity classification.Finally,a fine-grained opinion mining system for product evaluation is designed,and the evaluation element extraction algorithm and fine-grained sentiment polarity classification algorithm proposed in this paper are applied to the system,thereby realizing the function of product fine-grained opinion mining.