Research On Hybrid Recommender System Based On Deep Collaborative Filtering

As an important means to solve the problem of "information overload",recommender sys-tem is widely applied in the film community,social networks,bioinformatics,and other fields.The collaborative filtering recommendation has become one of the most widely used recommendation algorithms due to it is easy to implement and accurate recommen-dations.In recent years,although the recommender systems has played an increasingly important role in our lives,the quality of recommendation is not good enough.This is because these recommender systems lie in the intrinsic assumption that recommended users and items are independent and identically distributed(IID),the overall user and item characteristics and their non-IIDness have been overlooked,resulting in limited performance improve-ments.In reality,users or items are usually associated with rich coupling between users,between items,and between users and items,which can better explain the ways and rea-sons for users have personalized preferences to items.Non-IID recommender systems disclose the nature of recommendation and have shown its potential to improve the quality of recommendations and address the problem of sparse and cold-start.This paper mainly did the following work:(1)On the basis of NeuCF model,a neural collaborative filtering recommendation model based on side information-sideNeu is proposed by adding side information of users and items and considering the impact of the user and item content information on the model.Then,a comparison experiment with the NeuCF model was performed on the MovieLens dataset for different parameters.The results show that on the basis of independent and identical distributed learning,to some extent,the side information will improve the per-formance of the model,but the effect is limited.(2)Based on the Non-IID of recommender systems,this paper proposes a coupled rec-ommendation model based on deep collaborative filtering-sideDCF.The model firstly uses coupling models based on convolutional neural networks to concatenate side infor-mation with interaction information to learns the coupled feature vector and then make the concatenation with the implicit feature vector obtained by the deep collaborative fil-tering model.Then,we performed parameter sensitivity analysis experiments on three datasets containing side information.In addition,in order to explore the versatility of the sideDCF model,we also used bio-isomer network dataset for predicting new drug-target interactions.The results show that the model performs better than the current excellent model.