Research On Auction Method Based On Behavioral Game Theory

With the popularity of the Internet and the rapid development of e-commerce,auction,as a way of resource allocation and marketing mode,has attracted more and more Internet users and enterprises.Auction has become one of the research hotspots in the cross field of economic and computer science.Due to the diversity of auction items and the uncertainty of participants’ behavior,many new problems have emerged,such as insufficient assumptions of classical game theory and poor universality of the model.These problems can not be explained by traditional economic and game theory.Therefore,the research on the method of online auction combined with behavioral game theory and machine learning technology is of great significance to improve resource allocation and social welfare.The core problem of this dissertation is the behavioral game model and auction method of online auction.Taking keyword auctions and online item auctions as the application research background,the research is carried out based on behavioral game theory and machine learning technology.From the macro perspective,the multi-agent game model based on behavioral game theory is studied.From the micro perspective,the methods of bidding keyword matching and auction price prediction in multi-agent game are studied.At the same time,the proposed matching optimization method based on generalized second price mechanism,the method for predicting auction price based on Kalman filter and the matching and prediction optimization method based on behavioral game theory are applied to solve the problems of resource matching and price prediction.The main research contents and innovations of this dissertation are as follows:Firstly,a two-stage game model of multi-agent based on game theory is proposed to solve the problem of game and competition in the search engine market.In order to determine the auction price and resource allocation mode,the search engine companies are regarded as dynamic game agents under incomplete information.A two-stage game model of multi-agent and a method of analyzing equilibrium relationship is proposed.The simulation experiment shows that there is a Nash equilibrium with Pareto optimization in the second stage of multiagent game.In this equilibrium state,the game agents have no motivation to change bidding strategies.They achieve the global optimization.This method provides decision support for auction strategies.Secondly,in order to solve the problem of ranking and matching bidding keywords,improve auction efficiency and advertising quality,a matching optimization method based on generalized second price mechanism is proposed.A matching optimization method by calculating quality ensures the competition ratio.In order to further study the influence of quality score on matching performance,the role collaboration approach is introduced.A matching method based on the Environment-Class,Agent,Role,Group and Object(E-CARGO)model is proposed.It is applied to a ride-sharing system to solve the optimal matching problem of drivers and passengers.Finally,the simulation results show that the proposed method can effectively provide the optimal bidding strategies for game participants and improve the passenger carrying efficiency in a ride-sharing system.Thirdly,according to the heterogeneity and sparsity of auction data,high quality and quantity of training data is required in machine learning algorithms.A method for predicting auction price based on Kalman filter is porosed to solve the problem of low prediction accuracy.A predicting-updating dynamic iterative method based on Kalman filter is proposed to realize the dynamic prediction of online auction results.By dynamically optimizing the parameters,the problem of over fitting or low accuracy is solved when using multiple linear regression algorithm or support vector machine algorithm alone to predict the auction results in machine learning.The experimental results show that this method can effectively balance the performance of training data and prediction accuracy.The proposed prediction method can obtain higher accuracy with less training data,which is suitable for dynamic prediction.Finally,to improve the universality of matching and predicting methods,the utility theory is introduced into the auction.A matching and predicting optimization method based on behavioral game theory.is proposed.At the same time,the proposed matching method based on the E-CARGO model in ride-sharing is extended.A matching method based on behavioral game theory in ride-sharing is proposed.The experimental results show that the matching and predicting process considering utility is more in line with the behavior of game participants.This method improves the utility,prediction accuracy and carrying efficiency.