Research On Deep Learning Recommendation Algorithm Based On Heterogeneous Information Network

The recommender system can effectively solve the problem of information overload and is widely used in various fields.Collaborative filtering is one of the commonly used methods in recommender system.It is calculated by the user-item rating matrix.But the data is large and sparse in the recommender system,called the “data sparse”.The heterogeneous information network(HIN)can integrate information and can be used to integrate multiple types of objects and relationships in the recommender system.It can effectively solve the above problem.In addition,deep learning is designed to automatically learn the potential representation of useful data.It can solve complex tasks and reduce computational costs.Therefore,this thesis explores the recommendation problem based on HIN on the deep learning model.The main contents are as follows:1.We expand the original rating matrix and uses feedback information to form a new user-item interaction matrix.Then we used the meta-path to mine heterogeneous information.Finally,we added user attribute information or item attribute information.This thesis uses these three kinds of information as the input of the model to alleviate the data sparse problem.2.We proposed a model HINCF based on HIN on a Deep Neural Network.The HINCF can capture the relationship between non-linear users and items effectively.It achieves neural collaborative filtering based on heterogeneous networks and does the prediction rating.We used the above information as input to train the data in the model by minimizing the error between the predicted rating and the true rating.The model can automatically learn the potential vectors of users and items,as well as the interaction between them.Therefore,in the face of new users,a variety of input information makes the model complete effective recommendation,which effectively alleviates the cold start problem.We compared the existing algorithms with two well-known data sets called Yelp and MovieLens-1M.The results show that the proposed method shows good results and helps to improve the accuracy of recommendation.