It News Recommendation Method Based On Context Factors And User Interest Transfer

With the development of mobile Internet,more and more people frequently get external information from mobile phones.News APP is the most commonly used content acquisition APP in people's daily life.In the current environment,the speed of information transmission is coming.The faster the news content is more and more rich,the information overload phenomenon is serious,how to accurately locate the news that the user may like,and avoid the phenomenon of “information boudoir” to provide users with a wider range of news,which is the current news recommendation algorithm.The essential.Most of the previous news recommendation algorithms focus on the deep mining of news content,and the matching of news features with the potential features of users,while ignoring the dynamic changes of user interest on the timeline.The contextual attributes of users are related to news interests.Sexual mining.This study divides the news recommendation algorithm into two sub-problems,which are based on mining the interest migration phenomenon of users reading news,and exploring the relationship between user context factors and reading news interest hidden layer,establishing user interest migration learning model and context cross factor.Decomposition machine model.In the user interest migration learning model,the CNN method is used to extract the potential features of the news text on multiple time slices,and the user news interest changes reflected by the text features are mined,and the LSTM is used to predict the news interest of the user at the next time;In the cross-factoring machine model,considering the geographical location including the user's geographical location,the user's comment time period,the user equipment type.High-dimensional hidden layer feature extraction.Finally,the user interest migration learning model is combined with the context cross-factoring machine model,and joint training is performed to obtain the final prediction result for the news commented on the user's next time slice.In the course of the experiment,this study was trained on the real data of IT Home News APP.The model AUC index was higher than the Xgboost and Random Forest algorithms commonly used in industry,and achieved good results in practical recommendation.