Text Mining Based On Semantic Analysis,and Commodity Recommendation

In the commodity recommendation system based on the comment text,there are already many excellent recommendation algorithms and recommendation methods.But there is little analysis of what "buyers buy goods to give others".The situation that the buyer of this commodity is a person and that the user of the product is another person,can't simply use the word frequency analysis method to solve.So hereby,this paper presents a commodity review text mining method based on semantic analysis to solve this problem,intended to improve the existing commodity recommendation system,while finding a new way to solve the problem.This paper first collects the text of all the mobile phone reviews of Jingdong Mall,and prepares them by Word Segmentation,Part-of-speech Tagging and Semantic Role Labeling.Then,according to Valence grammar,Agentive(Commodity purchaser)and Objective(product user)is extracted from the preprocessed comment text.As the text of the commentary has the characteristics of colloquialism,there will be a large number of different forms of synonyms in the data.At the same time,it will also omit the use of the first person "I" to act as a guest,and the first person can't reflect the feature of the character,the need to be transformed into a third person form;so here to build "synonyms and the title of the self-proclaimed relationship control library" to unify the appellation and reasoning practice,and finally get "Agentive->verb->Subjective "form of structured data.Finally,all the characteristics of each commodity purchaser are counted;the average proportion is obtained;the commodity buyer's characteristic vector is constructed;and the commodity user's characteristic vector is constructed in the same way.Sorting the buyer's vector for all the goods,identifying the recommended product based on the purchaser's information,clustering all the merchandise user vectors,determining the recommended merchandise based on the user's information,and taking into account both factors to determine each other Recommended items.This paper implements the experimental program of the recommended system,and uses the collected data to carry out the product simulation experiment.Finally,the experimental results are analyzed and compared with the recommended results of other recommended methods.Qualitative analysis shows that the recommended method discovered the common characteristics of the purchaser and the user,and the recommendation is accurate and user-friendly.In comparison with the recommended results of "Lynx" and "Jingdong" mall,the similarity degree of 73.9%and 85.2%respectively reached,indicating that the recommendation system has certain practical value.