Research On Personalized Recommendation Algorithm Based On Local Sensitive Hashing And Latent Semantic Mode

With the rapid development of Internet,cloud computing and other information technologies,people’s demand for information is also increasing.However,the increasing amount of information on the Internet makes it difficult for people to get the information they want.Search engines and other information retrieval technologies relieve the pressure of information acquisition to a certain extent,but it is still difficult to meet the increasing needs of users.In order to enable users to obtain a better experience,recommendation system comes into being.The performance of the traditional collaborative filtering recommendation system is often affected by data sparsity,cold start and other problems.In addition,when processing large-scale data,the traditional recommendation system has problems such as high time overhead and poor real-time performance.Besides,most collaborative filtering recommendation algorithms mainly focus on improving the accuracy of recommendations,while ignoring the privacy protection of user information.In order to overcome the shortcomings of the existing recommendation system,this thesis does the following research work:(1)We propose a collaborative recommendation algorithm with privacy protection based on locality sensitive hashing(LSH)and matrix decomposition.In view of the traditional collaborative recommendation’s problem of high-time and privacy protection,We use the locality sensitive hashing technology,the user’s rating vector is converted to a binary vector,then the user is mapped to a bucket in a hash table.When searching similar users,we only need to find them inside the bucket.Because binary vector does not contain and can not backstepping user’s rating information,and the number of users in the bucket is far less than the total number of users,our method not only protects user’s information but also speeds up the search of the nearest neighbor in collaborative recommendation.To solve the problem of data sparsity,for the sparse rating matrix of users in the bucket,we used matrix factorization technology to fill in some missing rating values,and then carried out collaborative recommendation.Finally,experiments are carried out and our method is compared with other methods to verify its effectiveness.(2)In order to further solve the problems of data sparsity and cold start,and improve the accuracy of recommendation results,we fully mine user’s feature information,rating information and time-aware tag preference information,and design a hybrid collaborative recommendation algorithm based on multi-feature information fusion.Specifically,we mine the information of users’ preference for tags to find users with similar preferences.Since users’ preference for tags may change over time,we integrate the time factor into it and establish a user-tag-time tensor model.Considering the problem of data sparsity,we use CP decomposition technology to fill the sparse tensor.At the same time,we extract the user’s age,gender and other characteristic information,which can solve the problem of cold start to a certain extent.In terms of rating information,we take into account similarities between users and items.In order to improve the efficiency of recommendation and protect user privacy,local sensitive hashing technology is adopted before searching for similar users and items based on user preference and rating information.Finally,based on all aspects of information,we predict the rating for users from multiple angles,and the results are weighted together.And weight parameters can be got through learning and training.Through experiments,it is verified that our method can make accurate prediction and recommendation.