Research On Hybrid Recommendation Algorithm Via Denoising Autoencoder With Auxiliary Information

With the ever-growing dynamicity,complexity and volume of information available online,recommender systems have become an effective way to solve the problem of information overload.Collaborative filtering is one of the most popular technologies appiled in recommender systems.Traditional collaborative filtering only uses the rating matrices to make recommendation.But in general,the accuracy of recommendation is greatly reduced due to a large number of missing values in rating matrices and there is a problem of cold start for new users and items.In recent years,deep learning has been applied to recommender systems,but these problems still exist.To solve these problems,we utilize the effective representation in deep learning-denoising autoencoder and combine the auxiliary information to make prediction.The main contents of this thesis are as follows:1.A hybrid recommendation method via denoising autoencoder with feature information is proposed.The method considers the basic information of users and items at the same time.First,the users' preference are excavated from the ratings and the basic information of items.Second,we use the basic information and preferences of users as auxiliary information,integrate the ratings into denoising autoencoder focus on further learning from abundant auxiliary information and ratings to get the hidden feature representation of users,then reconstruct the rating matrices to make prediction.In the same way,the feature of items can be obtained from the ratings and the basic information of users.Experiments on Movielens-1M show that the integration of auxiliary information can improve the effectiveness of the recommendation.2.A hybrid recommendation method via denoising autoencoder with plot information is proposed.We first abandon the stop word and get word embedding of effective words using GloVe when dealing with the plot text,then get the embedding of each plot,the embedding contains semantic and grammatical information.The ratings and the embedding which are used as the auxiliary information are the input of denoising autoencoder.The best combination is found by changing the number of words in the interception of the plot and the dimension of word embedding.Experiments on Movielens-10 M show that our method performs better than other methods in the effective use of auxiliary information and performance improvement.