Research And Implementation Of Recommendation Algorithms Based On Situational Information

The traditional recommendation algorithms can effectively solve the problem of information overload caused by the rapid increase of information,but there are still problems,such as data sparsity and cold boot;In addition,traditional recommendation algorithms apply static recommendation to search users' long-term preferences,but it is difficult to find the changes of users' short-term interest preferences.Based on the in-depth study of the traditional recommendation algorithms,this paper applies some effective technical methods to study and improve it.Firstly,a Factorization Machine Recommender Algorithm based on the Context Information Transfer(CITR-FM)is proposed.The algorithm is based on the fact that the users' interest is easily affected by the surrounding environment.It makes full use of the context information to filter data,constructs the user model in a specific situation,and applies the transfer learning technology to transfer the selected samples.Combined with the matrix decomposition idea,it fully searches the user preferences in a specific situation,and finally generates recommendations.It can effectively solve the problems of low sample utilization,data sparsity and cold boot caused by context pre-filtering,and improve the accuracy of recommendation.Secondly,a Sequential Recommendation Algorithm Based on User Portrait(UPSR)is proposed.By constructing a sequence model,the algorithm pays more attention to users' short-term preferences and preference changes,makes full use of all the information collected to build user portraits,removes noise data,accurately finds users' partial relationship with items,and effectively finds users' shortterm preferences,thus produces real-time recommendation focusing on current situation.The experiment result shows that the recommendation algorithms proposed in this paper can improve the accuracy of recommendation and the coverage of recommendation list.While making full use of data samples and alleviating data sparsity,it can also make more accurate personalized recommendation to meet the different needs of users' long-term and short-term preferences.Finally,the proposed algorithms are applied to the specific recommendation system.After the interaction between the system and users,the specific recommendation list to meet different demands of users is generated,and displayed on the recommendation page that meet users' different needs,and then the user feedback information is recorded.The feasibility and rationality of the proposed recommendation algorithms are verified by system test and user record information.