Research On Commodity Recommendation Based On Viewpoint Mining

Customers are often required to write reviews for the products and associate services that they have purchased.E-commerce has been seen as a booming industry and the era has seen the increasing number of online shoppers and reviews.A hot item even has thousands of comments.This poses a difficult for both potential customers and merchants.One reason is that customers can hardly browse all reviews and make a wise decision in a short time.Another reason is that merchant are not able to keep track of the products and adjust their management.Accurate recommendation systems can help customer to find their needs.However,the current recommendation methods are mainly focused on user information and online behavior.Customer reviews can directly reflect product attributes and customer opinions.Thus,we consider to apply opinion mining technique to the recommendation system and proposed an opinion-based recommendation system.The main work of this paper includes:(1)The reviews should be preprocessed before extracting opinion words and feature words.The preprocess contains three steps,word segmentation,POS tagging and syntactic analysis.A small number of opinion words are manually selected and used as feeds for our model.Feature words are detected according to the co-occurrence of opinion words and feature words.We thus use clustering method to gain more opinion words and feature words.This is primarily based on the similar relationship in terms of the two words meaning and the propagation relationship between feature words and opinion words.The process terminates until no feature word or no opinion word can be identified.Using the above method,we can find the feature word sets and point words Set.(2)We use intuitionistic fuzzy set theory to make the polarity judgment of viewpoint words in this paper.After gaining the opinion word set,first of all,we can combine the word in a single clause with its star rating.Then,for complex sentence which contains more opinion words,the unknown polarity of opinion words is judged by intuitionistic fuzzy set theory and the known polarity of opinion words.(3)We add the comments to the recommendation model and proposed a hybrid recommendation based on the user goods and views in this paper.Features are extracted from user comment corpus according to the relationship among users,products and opinions.The theory of random forest algorithm and GBRT and LR is introduced to the hybrid model.Features are applied to the hybrid model.(4)The detail and the steps of our experiments are shown in the paper.The experiment with real e-commerce data is conducted to prove the result.The experimental results show that the proposed method can find the feature set and the viewpoint set effectively,and can better judge the polarity of the viewpoint words.Incorporating this method into a hybrid recommendation model can further improve the accuracy of recommendation.