Modelling Of Viewing Behavior Of Interactive Television And Mobile Apps And Application On User Program Recommendation

The advances in digital technology,communication technology and Internet technology have brought about tremendous changes in the mode of media communication.In addition to traditional methods such as digital television,terrestrial television and satellite television,interactive media such as interactive television,Internet viewing and mobile APP have been widely accepted by users for their rapid development.Under this trend,quickly grasping user preferences in massive programs to accurately recommend programs for users is of great significance for improving user-friendly media services.This thesis analyzes the difficulty of the traditional recommendation algorithm for program recommendation in interactive TV and mobile APP,then proposes the corresponding algorithm model,and validates the validity of the model on the real data set.Compared with traditional TVs,interactive TV can provide richer,more interactive on-demand services.The use of on-demand service viewers will generate a large number of interactive behavioral logs,then modeling on-demand TV viewing behavior can tap user interest to provide a basis for program recommendations.There are two main difficulties in the analysis of user behavior of interactive TV.First,TV ratings behavior is far less than Internet behavior.The strong sparsity of interactive viewing behavior data is not conducive to modeling research.Second,television reception is regarded as family behavior.The behavior log of family account reflects the coupling interest of many people and confuses the preferences of individual members.This thesis proposes a clustering coupled topic model to solve this problem.The model is used to couple the user's viewing clustering when viewing the interest topic modeling analysis,and to promote the optimization of the viewing topic modeling through user clustering.This paper further uses the results of the analysis of cluster-coupling theme model for user program recommendation.Experimental results show that the model can produce better recommendation results.Compared with TV viewing,users have higher demand for programs when viewing mobile APPs,and there are new difficulties in modeling the on-demand behavior of mobile APPs.On the one hand,APP viewing is facing the cold start problem brought about by a large number of new programs on the shelves.On the other hand,the vector expressions of user characteristics and program features obtained in the traditional recommendation model on implicit spaces are lack of interpretability.Therefore,this thesis proposes a label coupling collaborative topic regression model by making full use of user behavior data and program content data.This model integrates the tagprogram pair information to the user's click behavior and program text content coupling modeling,which alleviates the cold start problem and enhances the interpretability of the implicit spatial feature expression.The experimental results show that the model can better describe the user's viewing preferences,and in addition to recommending programs for users,the model has good performance in indicators such as Recall and MRR.