Research On Deep Learning Recommendation Method Based On Meta-path

With the rapid growth of information,recommendation methods have developed rapidly.By analyzing that the existing methods only consider explicit feedback information or implicit feedback information,it is necessary to combine explicit information with implicit information.Due to the excellent performance of deep neural network in mining non-linear feature relationships and potential information mined by different meta paths in heterogeneous information networks,this paper proposes a new meta path based personalized semantic recommendation method.Firstly,aiming at the problem of how to mine the potential information in features,this paper constructs an enhanced information network based on heterogeneous information network(HIN)by integrating the explicit information and implicit information in the recommendation system,so as to obtain more semantic feature representation.Specifically,the text(attribute)and score of user article constitute explicit information.The interaction between user and article forms implicit information by constructing meta paths,and then constructs enhanced explicit feature representation and enhanced implicit feature representation to mine potential information.Secondly,in order to mine the relationship information between features,this paper constructs the BFM-DNN model by combining the catboost factor machine algorithm fused with the decision tree(BFM)and the deep neural network(DNN).Specifically,the model generates enhanced explicit feature representation through feature extraction,while for interactive information,it selects different meta paths based on heterogeneous information networks to generate enhanced implicit feature representation,Then the bfm-dnn model is trained and predicted to obtain the list of recommended results.BFM algorithm combines features to mine the relationship between features;DNN uses neural network to mine the relationship of nonlinear features among feature data.Finally,based on the bfm-dnn model,a deep neural network recommendation method integrating meta paths is proposed.This method uses user item information to form enhanced explicit feature representation and enhanced implicit feature representation.It is trained by the bfm-dnn model,and finally the user recommendation sequence is obtained.The method in this paper is compared with the baseline method.Bfm-dnn significantly improves the evaluation index,which shows that this method has a good effect on mining potential information.