Research On Personalized Content Recommendation For Live Game Users Based On "User Profile"

How to extract valuable information from massive data and find out what users really need has become more and more important.In the game live broadcast industry,leading live broadcast companies are also paying more and more attention to the collection of user data,which includes a lot of valuable information.Through the analysis of such data,it is possible to deeply understand the behavior patterns and preferences of users,so as to improve the quality of live broadcast content and services,thereby increasing the revenue of enterprises.User profile technology is an analysis tool based on big data.It finds user characteristics by mining massive data,and then uses these characteristics to construct user profiles to obtain the "digital image" of each user.In this way,personalized content can be recommended for each user,content that may be of interest to each user can be recommended,and user satisfaction with using the product can be improved.To this end,this paper conducts research on personalized content recommendations for game live broadcast users based on "user profiles".First,using the literature analysis method,the knowledge system of user profile research and the research status of game live broadcast are sorted out from theoretical research to construction technology and application status,and the relationship between user profile and personalized content recommendation is expounded.Then in the data preprocessing stage,a data warehouse is established,and data mining technology is used to process the game live broadcast user data to generate user feature tags.Next,a personalized content recommendation system is proposed and designed.Then,based on the personalized content recommendation system framework and algorithm design,user profiles were constructed,and some user profiles were displayed and analyzed.In the stage of using user feature tags to generate user profiles,the author uses Embedding technology to solve the problem of sparse features,and uses the multi-bucket locality-sensitive hash algorithm to quickly and accurately search for the nearest neighbors of Embedding,and improve the recommendation algorithm.Finally,in the personalized content recommendation stage,the author studies and compares the recommendation effects of deep learning algorithm models such as GBDT model,LR model,FM model,and Deep FM model.In view of the corresponding advantages and disadvantages of each algorithm model,The author tried to integrate Deep FM and LR models for recommendations,and found that this recommendation model can achieve more accurate recommendation results.In addition,the accuracy of the recommendation results was verified online in the overseas M project of a leading game live broadcast company in China.The main conclusion of this paper is that user profile technology can be well applied to the field of game live broadcast,and the game live broadcast content recommended by the fusion of Deep FM and LR models in the M project is more popular among users.Compared with the traditional game live broadcast content that does not use a recommendation algorithm,the user’s click rate and stay time have been significantly improved.