Research And Implementation Of Heterogeneous Network Recommendation Algorithm Based On DNN

Mainstream recommendation algorithms have a large demand for interactive data between users and projects,have a serious problem of cold start,and cannot make full use of auxiliary information.Heterogeneous networks are closer to the network structure in real life,and semantic relationship mining of rich node and edge information can improve the accuracy of the recommendation algorithm and effectively alleviate the problem of cold start.The recommendation algorithm based on heterogeneous network has become a research hotspot.The existing recommendation algorithm has the following problems :(1)when the user's historical interaction data is rare,the recommendation algorithm will have the problem of cold start,and the prediction accuracy will decrease;(2)the semantic importance of auxiliary information to the implementation of user project interaction is not paid attention to;(3)Shallow neural network is used to learn the recommendation relationship between user features and item features,but the deep relationship cannot be learned.Therefore,this paper proposes DHRec,a Heterogeneous Network Recommendation algorithmbased on DNN,and applies it to the constructed tourism Heterogeneous Network dataset.The specific work of the paper is as follows:(1)DHRecIn order to overcome the problem of cold start when there are few interactive items,heterogeneous network representation learning is integrated into the recommendation algorithm.Rich semantic information and attribute information are integrated into node representation through heterogeneous network representation learning.In order to improve the algorithm's learning of the semantic contribution of auxiliary information to click behavior,the attentional mechanism is used to weight and fuse the representations under different meta paths.The attention mechanism is used to learn the influence of different meta paths on the user's selection of items,and the user's behavioral preference for different semantic meta paths is fully captured.To solve the problem that shallow neural network cannot learn the deep relationship between user characteristics and item characteristics,deep neural network is used to fit the complex nonlinear relationship among user,item and interest representation,and extract deep interactive information.DHRec was validated against standard data sets Yelp and Movielens.Compared with previous models,experimental results showed that AUC increased by 2.3% to 3.5%,and DHRec also performed well in alleviating cold start problems.(2)Construct tourism heterogeneous network datasetCold start is a common problem in tourism recommendation,but there is no open tourism heterogeneous network dataset.In this paper,the crawler technology is used to obtain the tourist information of scenic spots on Ctrip,and data cleaning,keyword extraction,data sorting and other operations are carried out to construct the tourism heterogeneous network dataset.(3)Personalized travel recommendation based on DHRecWhen DHRec was used to make personalized tourism recommendation for Shijiazhuang on the tourism heterogeneous network,Compared with previous models the AUC value increased by about 2.2% and the GAUC value increased by about 4.5%.