Research On Personalized Recommendation Algorithm Based On Emotional Analysis

The personalized recommendation algorithm is to present the user's potential demand products to the user by filtering and filtering the massive information.Among them,collaborative filtering algorithm is the most widely used and most successful personalized recommendation algorithm in e-commerce and social networks.It is based on the user's scoring records or the invisible data generated by the user's browsing web pages to mine user's preferences,there by recommending users' interesting product.However,there are obvious drawbacks in this type of recommendation.The historical behavior of the two users on the product as a whole is similar,but their preference for certain attributes of the product is different,which obviously restricts the recommendation effect of a recommendation system.In online shopping surveys of users,users find that when they purchase a product,they pay more attention to the purchased or used user's evaluation of the product because the evaluation information of the product contains the subjective opinions,emotions,and preferences of the user on certain attributes of the product.Therefore,it is of great significance to construct a more accurate and perfect recommendation system based on the product comment information mining product attribute words and the user's emotional tendency analysis of product attributes.This thesis aims at the review information of Mei Chang.com.cn as a research object,and proposes a personalized recommendation algorithm based on sentiment analysis,which includes the extraction,classification and emotional orientation analysis of product feature words and the construction of user's interest model,etc.Focuses on research and analysis.(1)The features are extracted from the review text based on two aspects of frequent features and infrequent features.Firstly,Apriori association rule algorithm is used to extract frequent features,and frequent features are filtered by the co-occurrence rate of point mutual information(PMI)and feature-emotion words.Then,the infrequent features are extracted for the characteristics of co-occurrence of emotional and infrequent features with frequent features.(2)Taking the string,semantic similarity and the co-occurrence degree of feature words and opinion words as the association weights among feature words,the K-means initial centroid selection was implemented to realize feature automatic clustering.(3)Fine-grained analysis of gourmet restaurant reviews,using emotional word matching methods,judged the polarity of the restaurant's characteristic sentences.The user's interest model is constructed from the user's attention to the characteristics of the gastronomy shop and the degree of pickiness.The research of this thesis abandons the recommendation of single factor for user rating.Through the multi-dimension extraction of culinary shop review features,the multi-attribute factors of culinary shop are taken into consideration,which avoids recommending those merchants who get high marks through bidding ranking,speculation,etc.;At the same time,the users' comments were analyzed emotionally,and the potential inner activities of the users were analyzed in depth,and their interest in the nature of the gourmet restaurants was tapped.Theoretical and experimental results show that the research of this thesis greatly improves the performance and effect of the recommendation compared with the traditional collaborative filtering algorithm.At the same time,the recommendation of popular articles is reduced,which is more in line with the requirements of users.