Research On Customer Opinion Mining Based On Domain Knowledge Base

With the development of the Internet,e-commerce has had a profound impact on people's living habits.Online ordering,online shopping and other consumption methods have become an important part of modern people's lives.What followed is the huge amount of commentary information published by consumers on the shopping platform.What followed is the huge amount of commentary information published by consumers on the shopping platform.This information seems to be confusing,but it has great research value.Through in-depth analysis of these commentary information,mining valuable information,can provide information support for the rational decision-making of enterprises or consumers.For enterprises,opinion information can help business managers monitor the quality of products and services,so that they can find their own problems and improve them in time to improve their competitiveness.For consumers,valuable opinions can play a role in consumer guidance.Therefore,it is necessary to conduct opinion mining research on the review text information on the e-commerce platform.In reality,users often have different needs when they get opinions,such as opinions on the user's needs for several major aspects of the product,or opinions on a specific attribute of the product,so the Coarse-grained Opinion Mining has not been able to effectively meet user needs.To address the above issues,this paper builds a domain knowledge base based on the commentary text corpus,using natural language processing technology and statistical-based methods,and introduces the domain knowledge base into the opinion mining research.The domain knowledge base is constructed to find the entity(aspect term)in the comment text that can be mapped into the knowledge base,and then extract the feature viewpoint sentence.Then,we extracted feature tuple from each feature viewpoint sentence,and analyzed emotional polarity of every aspect term,so the feature-emotion set is obtained.Finally,the feature-emotional aggregation is realized through the semantic relationship network in the domain knowledge base.On this basis,the feature-emotion map is constructed according to the different needs of the user,and the opinion information in the comment text is presented in a concise and clear way.The main work of this paper is as follows.In order to ensure the accuracy of word frequency statistics,we use the NLTK tool library for part-of-speech tagging and stem extraction.On this basis,the domain membership degree analysis method based on the TF-IDF principle is used to extract the evaluation objects from the text corpus to form a collection of terms.Then,using the Pointwise Mutual Information(PMI)method to extract the relationship between the terms,form a semantic relationship network,and build a domain knowledge base finally.(2)After the comment text is segmented to long sentence,it is further divided into several short sentences,so that only one term is described in the short sentence.This paper presents a combination of rules and word2 vec model to construct a synonym set to ensure that the term is mapped to the domain knowledge base.We name this sentence as a “feature-viewpoint sentence” while the match is successful.In addition,this method also ensures the accuracy of feature-emotional aggregation in subsequent work.(3)We extract the feature sentiment triad from the feature-viewpoint sentence,and combine the basic sentiment dictionary SentiWordNet with the negative word dictionary to form the final dictionary.And then use the dictionary-based method to analyze the emotions of the evaluation objects in the feature emotion triples and determine the emotional polarity.The result shows the accuracy rate is 73.6% in the manually labeled data,which verifies the effectiveness of the emotional polarity discrimination method.(4)According to the different needs of users,this paper uses the semantic relationship network in the domain knowledge base to realize the aggregation of feature emotions.Finally,the feature-emotion map is constructed,and the opinion information in the comment text is displayed in the form of graphs.