Research On Integrating Social Information For Recommendation Algorithm

With the popularity of the Internet and the rapid development of big data,it is increasingly difficult for people to obtain effective information from massive data,so the demand for personalized recommendation is also increasing.Based on the development of mainstream social media and the dynamic interests and preferences of users,this paper studies and analyzes the data sparsity problem in the recommendation system,and proposes solutions:By introducing the time information of user behavior and cross-system social information,this paper proposes a collaborative filtering recommendation model that integrates social information and time effect.First,LSTM network is used to model users' short-term interest preference,integrate users' long-term and short-term interest preference as users' real interest preference and user attributes to extract user characteristics,and use project attributes to extract project characteristics.Then,the semi-supervised learning method based on graph is adopted to model social information.The updated user representation is taken as the input of the next training based on the recommendation model through the user representation function learned on the social network.Finally,the interactive prediction score of the user and the project is obtained and the recommendation list is generated.In order to prevent over-fitting in the training process,this paper proposes a global loss function regularization method,and constructs a triple constraint loss function solution model integrating user embedded regular terms,item embedded regular terms and attribute embedded regular terms.From the results,it can be found that model STCF performs better than others benchmark algorithms on the evaluation metrics,The research conclusion of this paper shows that this model is feasible and effective in solving the problem of data sparsity in the field of tourism and books.Compared with other popular recommendation models,the experiment is improved on the two different data sets respectively compared with the Dynamic Graph Recommendation Model by [2.3%,20.1%,22.8%]and [1.9%,16.5%,12.1%] respectively in metrics of the AUC,recall rate and precision rate.It is verified that the model achieves certain improvements.