Research On A Recommendation Algorithm About Combining Sentiment Analysis With Fuzzy Kano Model

E-commerce plays an increasingly important role in people's daily lives.People are more and more accustomed to online shopping.It is becoming more and more easy for researchers to obtain subjective opinion data of users such as online comments.However,due to the explosive growth of trading volume,the information overload is also generated while retaining a large amount of mass data.In the face of massive data,it is particularly important to establish an efficient and effective filtering mechanism to discover the real demand,so people pay more attention to the research of recommendation system.Recommendation system is a mechanism that utilizes the history data to provide suggestions which assist people in choosing products.Personalized recommendation is to customize the information and products of interest according to certain characteristics and behaviors of users.Collaborative filtering is a classic algorithm of recommendation system,which is divided into two categories: user-based collaborative filtering(User CF)and item-based collaborative filtering(Item CF).However,traditional collaborative filtering is based on user rating as a measure of similarity.It does not take into account the fine-grained feature analysis of users or products.It can only contain the holistic attitude of users to products,but not the personalized factors of users and the uniqueness of the product in some way.The product reviews often include the user's product preference and sentiment polarity,to a greater extent,which will affect the purchase behavior of the user.So the method of using content-based opinion mining can be more than the score-based method,and it also have better precision.Although the current ItemCF algorithm research has turned to opinion of product features,it is only a simple study of the similarity between products,but few researchers combine the relationship between people and items.In other words,The relationship between user needs and certain features is need to be considered.This relationship is reflected in the measurement of the similarity of the items,which needs to select the features that the user really needs rather than all the features.The lack of such considerations also affects the performance of the recommendation system.Another problem is that the recommendation algorithm has not interpretability.In the past,the recommendation system was to map users and products to vector space to calculate the spatial distance to obtain similarity for recommendation simply.It can only be explained by mathematical but without the factors of the actual scene.The Kano model's demand grouping ability can solve these two problems.Based on this background,this paper has done the following three works:1)This paper review the literatures in three research field: recommendation system,sentiment analysis and Kano model.The part of recommended system introduces in the detail the principle,research status,common problems and solutions of the collaborative filtering algorithm.The part of sentiment analysis reviews the specific research problems at the chapter level,sentence level and word level,and the current existing results.The part of Kano model introduces the principles and application methods,the fields in which the Kano model is applied and the current research status detailly.2)To solve the above problems,this paper combine the Kano model with feature sentiment grouping based on opinion mining.The algorithm uses the sentiment analysis to calculate the sentiment polarity from the comment data,and then combines the fuzzy Kano model to classify the total features according to the degree of demand,then chooses the high-demand feature to replace the total features to metric product similarity.On the one hand,it alleviates the sparseness of collaborative filtering,on the other hand,it improves the efficiency and effectiveness of the recommendation system.3)The method of measure similarity proposed in this paper is based on the individual needs of the user.The similarity calculation is performed in a way of scene modeling.The use of the Kano model also makes the recommendation system explanatory.This paper proposed algorithm model is applied to the actual life.It choose the part of smartphones and musical instruments categories of the Amazon open source dataset to experiment.The final experimental results show that our proposed ASCF algorithm batter than the traditional collaborative filtering in precision and effect.At the same time,to prove the rationality of the Kano model applied in this field,this paper conduct relevant empirical research.