Analysis And Research On Information Interaction Of Live Streaming Platform

In recent years,China's live streaming platforms have developed rapidly.Live streaming has been leading the industry in terms of the size of user groups,the richness of live streaming content and scenes,as well as the favor of the capital market and the application of intelligent technologies.However,with the rapid development,Live streaming platforms also encounter problems and challenges such as homogenization of live streaming content,insufficient liquidity and impact of short video platforms.Therefore,it is necessary to analyze the behavior characteristics of anchors and audiences by researching the information interaction process between users of live streaming platforms,which is not only conducive to better understanding the operation and development mode of live streaming platforms,but also conducive to solving the problems and challenges encountered of various live streaming platforms.The main contents are as follows:(1)This paper takes Longzhu live streaming platform as the research object,collects data from the platform and makes a statistical analysis.It mainly includes the variation and distribution characteristics of the live streaming situation of anchors,the barrage,gifts and popularity value of live streamers;the distribution characteristics of audience's barrage-sending and gifts-tipping behaviors;the differences of user information interaction between games,entertainment and sports live streaming types.In addition,the influence of anchors' gender distribution;the distribution of barrage length;audience's preference for gifts of different types of live streaming content;and the influence of whether anchors join the guild on the number of living days,the number of barrage in the live streaming room,the number of gifts and the popularity value are also analyzed.Through the above statistics,we have a comprehensive understanding of live streaming characteristics of Longzhu live streaming platform and the information interaction between anchors and audiences.(2)Clustering analysis is conducted on features built for each "Highly engaged" audience who sends more barrage and more gifts on live streaming platform to mine the characteristics of user behaviors.First,we build multidimensional features for each audience,and then K-prototypes algorithm is used to cluster them into three clusters with obvious heterogeneity:ordinary audience,engaged audience and super engaged audience.Suggestions are provided for the live streaming platform to mine the potential value of three types of users according to their respective characteristics.(3)This paper proposes a method of popularity value prediction.According to the prediction model framework,we build multidimensional features for each live streaming room after data preprocessing,then the Random Forest,XGBoost,LightGBM algorithms and the different combination of features are conducted to predict the popularity value of each room per hour.The results of experiments show that the LightGBM2 model has the most minimum MSLE,the best fitting with the actual value,and the least time consuming for model training.The prediction of popularity value can be done well with the LightGBM2 model.