Research On Cheat Warning Model Based On User Behavior Analysis

With the concept of "data drive production" deeply rooted in the hearts of people,the data have become the motive force of social life,industrial production and enterprise operation.As of March 2017,the Ministry of Education has approved 35 universities to establish the major of "data science and bid data technology".Under the direction of this trend,the big data competition has become a modern form of data mining.The competition is held by the enterprise itself,or the professional platform.The teams will gathered in the competition and find the algorithm optimization scheme or the problem solution within a limited time.However,in the process of competition,there will be some controversial issues on fairness.This puts forward higher requirements for the operation team,creating a fair competition environment and evaluation mechanism,preventing loopholes,and making up in time when finding loopholes.This dissertation mainly studies the cheating behavior in the competition platform based on the user behavior,the feature extraction is carried out and some models are used to be trained.The dissertation focuses on the access behavior of the platform users,and uses all kinds of indicators to conduct behavior definition.It identifies the outliers in user access behavior by anomaly detection.Then combined with the similarity of the user's file,the early warning of cheating will be analyzed.The main research work is divided into the following aspects:1.From the perspective of user access,the user browsing behavior data are analyzed.When browsing web pages,users have common purpose behind them,that is,to find the contents of competitions,participate and submit the results of competitions.Therefore,these normal user accessing behaviors may have similar behavior patterns.Abnormal users usually have special purposes when they use platforms,such as to get more submission opportunities and platform feedback.Thus abnormal users behave differently based on certain dimensions.This dissertation analyzes the data that can describe user browsing behavior and distinguishes two types of users with strong related features.2.This dissertation analyzes the user's submitted documents and designs a similarity test method for the characteristics of the submitted documents.Meanwhile,a detection method based on the change trend of the performance is proposed according to the corresponding performance sequences.In this dissertation,a whole directional test on the file itself,the time of file submission,the score of files are made according to the previous submission of different users and the previous submission of the same user.3.The early warning method of cheating is researched based on the whole-directional behavior of the users.According to the data of users' access behavior and the data of submission behavior,this method can get through it and process it into regular user behavior characteristic data under a particular competition.The user feature is used in combination with file similarity test to improve the cheating warning model and establish the automatic handling mechanism of platform.4.The model described in this dissertation has been implemented and tested on the competition.The experimental results show that the average prediction accuracy of user behavior under svm and xgboost algorithm is 95%,and both can predict steadily under multiple competitions,which shows that the model is effective and universal.In this dissertation,the optimal opinion on the final output of the model is given.In addition to the effectiveness and universality of verification,cheating warning model helps the real operation in the real business environment.The complaint from user is less,which has reached the original intention of the model design.The model proposed in this dissertation can improve the fairness of competition and the operating service level to a certain extent.Nowadays,whether improving the competition mechanism,creating a fair competition environment,or improving the precision of the service level can not be separated from the support of data.In this dissertation,the model is used to analyze all the data of the user and is applied to the early warning of cheating.In the future,the model can not only adapt to the changing user environment through feature optimization,but also expand the application of user behavior analysis method to personalized competition operation,operation mode exploration,operation effect evaluation and some other fields.