Research And Application Of Hybrid Recommendation Algorithm Basedon User Behavior And Item Content

With the advent of the information age and the popularity and development of the Internet,overwhelming network information has brought us into an era of information overload.Accompanied by information technology skyrocketing,a user can’t timely and accurately in the network find the real information needs.In this context,from the original knowledge classification technology change to the recommendation technology is inevitable.Recommendation system does not like the search engine system that accepts the user’s explicit needs to search for users but dig out the user’s potential interest and provide recommendation for user’s according to the user’s history behavior records.In the research field of recommendation algorithm,the main research point is how to solve or alleviate the problem of data sparsity,cold start,and low quality of recommendation.Collaborative filtering algorithm can’t effectively solve the problem of cold start of items,which is based on the user’s behavior information and content-based recommendation algorithm always lead to poor diversity.This paper combines user behavior and item content to propose a hybrid recommendation algorithm based on project topic and a hybrid recommendation algorithm based on item type deletion.The two algorithms proposed in this paper can solve the problem of cold start and poor quality.The main contents of this paper are as follows:1.Combined with item type information and user behavior information,this paper proposes a hybrid recommendation algorithm.Firstly,using the topic model LDA to extract the topic vector from the item type information.Secondly,selecting several representative dimensions to replace the entire topic vector according to the characteristics of the topic vector of each item.Third,using the item topic vector calculate the similarity between items and embed the calculated items similarity into collaborative filtering algorithm to establish the recommendation model.Finally,use the established recommendation model to recommend item list for users.The results show that this method can solve or alleviate the problem of cold start.Compared with the classical cooperative filtering algorithm,the proposed method has a large improvement in recommendation accuracy,diversity and coverage.2.Combined with item introduction information and user behavior information,this paper proposes a hybrid recommendation algorithm.Firstly,using the chi square test to select several characteristic words as feature attributes from the item text key word set for each item type.Secondly,in the calculation of the eigenvalues of each characteristic attribute,the Word2 Vec tool is used to train the item text information corpus to get the word vector of each word,and calculate the similarity between words using word vector.If the two words are similar enough,the two words are assigned to the same attribute.Then calculate the eigenvalue to solve the problem of data sparse.Third,use type as the target of eigenvector to train classification model for each type,and then use the trained model to predict whether an item contains the type.Fourth,merge the predicted item types and the types of item contained in the content of item.Fifth,using LDA model to extract the topic vector of item from the item-type information and selecting several representative dimensions to replace the entire topic vector according to the characteristics of the topic vector of each item.Sixth,using item types to calculate the similarity between the items,and embed calculated similarity into collaborative filtering recommendation model to establish the new recommendation system and generates a recommended list for a user.The experimental results show that there is a significant progress in the accuracy and coverage when compared with classic collaborative filtering algorithm.