User Opinion Extraction And Pain Point Recognition Based On Dependency Syntax And Product Feature Library

With the rapid development of online shopping,many online shopping platforms have accumulated a large number of user comment data.How to effectively analyze and use these user comment data,exploring the value of massive user comment data,and provide reference for related practice,which has important theoretical and practical significance.In this context,the user opinion mining emerged.The research on user opinion mining can be generally divided into two categories: opinion extraction and opinion sentiment analysis.Opinion extraction is the premise and basis of the opinion sentiment analysis,and they complement each other,which are collectively called the “user opinion mining”.The research on opinion extraction can be divided into two parts: manual extraction method based on rules and patterns and automatic method based on machine learning and deep learning.Although the performance of automatic method is better than that of manual method,the model training depends on a large number of human labeled corpus,and the quality of manual annotation greatly affects the model's performance.In addition,the interpretability of the extraction results is also weaker than the manual methods based on rules and patterns.This paper uses dependency syntactic analysis technology and product feature library based on Word2 Vec word vector technology to extract the user's opinion,and improve the accuracy and quality of opinion extraction by introducing the part of speech of the word pairs,dependency combination and dependency distance constraints of dependency pairs.The research on opinion sentiment analysis can also be divided into two categories: artificial method based on sentiment dictionary and automatic method based on machine learning and deep learning,which all have achieved good results.However,these researches mainly classified the opinion sentiment as positive and negative or classified the sentiment as the positive,neutral and negative.In the reality of limited resources and diverse user needs,the research results' practicability will be weakened to a certain extent.This paper multi-dimensionally aggregates the user's opinion and analyses their emotional tendency and intensity based on the results of the proposed opinion extraction.Finally,the user attention index is introduced to comprehensively calculate and visualize the score of the user pain points in each coarse-grained attribute or fine-grained feature,which is beneficial for the related enterprises to comprehensively and deeply grasp the users' pain points and needs,and provides a guidance for the product's upgrading and improvement.The experimental results on Xiaomi 9 mobile phone comment data show that the user opinion extraction method based on dependency syntax and product feature library proposed in this paper has a better performance than the nearest distance method and SBV polarity transfer method,and there is greater improvement in accuracy,recall rate and F1 value than the two basic methods,which proves the effectiveness of the method proposed in this paper.And the consistency between the results of user pain point recognition method and the proportion of poor evaluation in the comment corpus,also verifies the operability and effectiveness of the user pain point recognition method proposed in this paper.