Application Of Behavior Sequence Pattern Recognition In User Interest Discovery Of Pan-Entertainment Platform

Pan-entertainment is one of the important features of Internet development at this stage.The trend of mobile Internet entertainment has activated the inherent and exciting entertainment social needs of users.Users generate a large number of behavior logs during the use of the product,including information on the time,location,content,interaction,consumption,etc.How to make full use of the massive logs left by users on their products,using data mining technology to get deep insights of users' preferences,extracting valuable information for the company,and maximizing platform benefits have become a hot topic in industry and academia.Behavioral logs with timing information,also known as behavioral paths,contain valuable information such as user jump habits and path preferences.In order to guide users to continuously experience the value of products and maximize user interest,the Internet industry generally adopts user behavior path analysis method,which can be called behavior sequence pattern mining,which is a combination of sequence mining technology and user behavior path,exploring behavior law,discovering interest patterns,so as to know the product usage and find the optimization direction method.This paper actively explores the application of behavioral pattern recognition technology in user interest discovery.From the perspectives of mining preference path,extracting key function points and analyzing low-active user characteristics,a multi-level user interest analysis system is constructed.Based on the real behavior data of the users of the entertainment platform,the feasibility of the system is demonstrated through experiments.The specific work is as follows.Firstly,the concept of effective behavior and effective sequence is put forward,and the data cleaning and sorting are completed accordingly,and the basic data for subsequent analysis is formed.Then the basic data exploration and feature analysis are carried out from the perspectives of sequence length and behavior type.Secondly,based on the parameter estimation problem of Hidden Markov Model,experiment of user interest pattern mining is conducted.The experiment groups the data according to the length of the sequence.In order to ensure that the model training scientifically and is not misled by the data with too large differences in distribution,this paper also proposes a hierarchical sampling scheme.In the result evaluation stage,based on the fact that the scene to which the known behavior belongs naturally belongs to the hidden state,an indicator of the hidden state accuracy rate is designed.By using the accuracy,a better model is selected,and the model is used to simulate the most probable behavior path.The path preference of the user is found in these simulated paths,which proves that the model through the real sequence can maintain good fitting accuracy.Finally,the practical guiding significance of these simulation sequences in the discovery of interest is explained.Then,in order to solve the problem of discovering important functions in the pan-entertainment platform,this paper proposes a scheme of constructing a graph with behavior as nodes and a jump relationship among behaviors as edges based on graph theory in the field of mathematics.The key nodes in the problem are found.Among the solutions of discovering key nodes,three indicators that characterize the importance of nodes from different angles are selected.In the experimental analysis stage,using the grouping data of different sequence lengths,the graph construction and the key nodes are respectively found,and the head results are obtained under each index.By comparing the results of different groups and indicators,The functions that plays an important role in the system is given,and the optimization suggestions for the product are proposed for this result,which proves the feasibility of the graph model.Finally,focusing on the issue of pan-entertainment ecological promotion,a comparative experiment was designed to compare the behavioral sequences of low-active users and highly active users of the platform,with high-activity data indicators as reference,and observation indicators for interface bounce rate and interface exit rate,the difference ratio of the indicators used to measure the difference between the two groups of data is proposed.For the bounce rate and the exit rate,the behavior interface with the proportional difference greater than the threshold is selected as the experimental result.Then,from the actual business,it explains the specific impact of these interfaces on user churn,and gives the optimization plan.Finally,it also proposes other targeted promotion suggestions.