Research On Behavior Analysis Of Recommendation System Based On Computational Experiments

With the wide application of Internet technology,data and information show an explosive growth trend.To solve the problem of information overload,recommendation system has been widely used.The application of recommendation systems effectively reduces the decision burden of users,however,there is a potential risk of influencing user behavior and perception due to the personalized distribution process of their internal algorithmic models.Based on this,the research on the behavior of recommendation systems has become a research hotspot in academia and industry.Researchers at this stage have mostly focused on how to improve the performance of recommendation systems by learning user data to obtain higher click through rates,but they seldom study the mechanism of the behavior of the recommendation system affecting user selection or cognition.At present,this kind of research mainly has three shortcomings:(1)the behavior analysis of recommendation system is limited to qualitative description,but there is no quantitative analysis of behavior process and behavior result design indicators;(2)Pay more attention to the changes of users’ dynamic behavior,ignore the dynamics of recommendation strategy of recommendation system and the circular feedback between recommendation system and users;(3)Unable to effectively deal with the interference caused by noise.These problems directly lead to the difficulty for researchers to effectively predict and supervise the behavior of the recommendation system.In order to solve the above problems,this thesis introduces the means of computational experiment,takes advantage of its accuracy,controllability,simple operation and repeatability to put forward the analysis and research method of recommendation system behavior based on computational experiments,specifically including: taking the process of recommendation system behavior affecting the diversity of user selection results as the research goal,define and refine the roles of different recommendation system agents and user agents according to the participants and interaction logic in the online interaction process,and accurately describe the interaction process between them;Through the computational experimental method to simulate the online interactive evolution,define the content diversity index,dynamically measure the main behavior and behavior result change of the recommendation system agent,and analyze the action mechanism of the main behavior of the recommendation system agent.The behavior analysis method of recommendation systems based on computational experiments effectively isolates noise from the online interaction process.The method intuitively analyzes the dynamic co-evolution between the recommendation system and user selection.Based on the existing research results,this thesis proves the feasibility and effectiveness of the experimental method.The experimental results show that it is an effective method to analyze the behavior of recommendation systems through computational experiments.The research work of this thesis provides research ideas for the development of machine behavior.Moreover,it has important reference significance for the governance of recommendation system platforms and the healthy and benign development of recommendation systems.