Research On Personalized Recommendation Algorithm Based On Text Tag

With the rapid development of the Internet and the rapid popularity of e-commerce,huge amounts of information are flooded in people's lives.The problem of information overload is becoming more and more serious.In order to solve this problem,the recommendation system was born.It analyzes the user's historical behavior record to draw the user's interests and hobbies,and then extracts the content that users may be interested in from massive information.As the most critical component of recommendation system,recommendation algorithm has received extensive attention and research in recent years.Aiming at the problems of sparsity and cold start in the recommendation algorithm,this thesis studies from two aspects: data source and recommendation algorithm model.The main work completed in this thesis is as follows:After in-depth analysis of the existing collaborative filtering recommendation algorithms,this thesis proposes to introduce text tags into the recommendation algorithm.The classic collaborative filtering recommendation algorithm usually uses rating data to train recommendation model,however,the recommendation system includes data such as comments and tags in addition to the rating data.In view of the strong correlation between text tags and the items to be recommended,reasonable use of this relationship can increase the accuracy of the calculation when recommending items to users.In this thesis,all the text tags are extracted from the history behavior records of the user and the tag space is constructed.Then,according to the membership relationship between the items and these tags,the vector representation of each item in the tag space is obtained.Finally,the tag features of each item are extracted with the deep learning method,that is,using the tag vectors of all items as input to train an autoencoder neural network with a single hidden layer.Another work of this thesis is to study the operation principle of matrix factorization model and neighborhood model,and propose a matrix factorization recommendation algorithm based on text tag.The working principle of the neighborhood model determines that it can capture the local information better,and the matrix factorization model can better capture the global information by modeling the potential feature vectors of the user and the item on the known rating set.The two models are essentially complementary.In this thesis,a suitable integration framework is used to integrate the characteristics of them.According to the tag characteristics of each item extracted from the text tag in advance,the near neighbor set of the target item is determined.The item vectors in the prediction model are extended by using the tag features of the item in the set.Experimental results show that compared with the traditional collaborative filtering algorithm,the accuracy of the proposed algorithm is higher.