User Abnormal Viewing Behavior Detection Method Based On Improved Canopy-FCM And Isolated Forest Algorithm

In recent years,with the rapid development of Internet technology,IPTV has been favored by more and more home users,which has led to rapid expansion of IPTV services and continuous growth in the scale of users.With such a rapid development speed,IPTV services should not only focus on the improvement of information transmission speed at present,but should pay more attention to the user experience when enjoying IPTV services.In order to improve the user experience and provide decision-making suggestions for the video recommendation and advertisement push services provided by the service provider,it is particularly important to be able to accurately analyze the user's viewing behavior.However,due to the large scale of ratings data,it is inevitable that abnormal data will appear in the process from generation to collection.Therefore,it is of great significance to detect abnormalities in ratings data.Based on the above problems,this thesis proposes a method for user abnormal viewing behavior detection based on improved Canopy-FCM and isolated forest algorithm.The main research content of this article is reflected in the following aspects:1.Propose an improved Canopy-FCM clustering algorithm.Combining the Canopy algorithm and the FCM algorithm,first use the Canopy algorithm to roughly cluster the data set,and use the number of clusters and cluster centers as the input of the FCM algorithm to perform more efficient and accurate clustering.Based on the fuzzy decision theory,the selection of the fuzzy weighting index m of the FCM algorithm is determined.Based on the maximum and minimum criteria and the maximum density rule,the new Canopy selection and setting process in the Canopy algorithm is improved.2.Propose an improved isolated forest algorithm.In the training phase of the isolated forest,a cutting point selection algorithm is proposed,which can make the position of the data point on the separation tree more accurate.In order to comprehensively consider the anomaly degree of all the features of the sample data,this thesis proposes a comprehensive anomaly score based on the combined weight,and uses the entropy weight method to determine the size of each weight in the combined function.In the phase of abnormal analysis,an algorithm to determine the abnormal score threshold is proposed,which can more intuitively distinguish abnormal data samples from normal data samples.In order to verify the superiority of the improved clustering algorithm and anomaly detection algorithm proposed in this article,this article uses user ratings data and related public data sets to compare experiments with other traditional clustering algorithms and anomaly detection algorithms,and uses related The evaluation index is used to evaluate the experimental results,which verifies the accuracy and effectiveness of the improved algorithm proposed in this thesis.