Research Of User Behavior Analysis Based On Machine Learning

With the coming era of mobile internet and the popularity of mobile intelligent devices, traditional telecom business models and the services users can obtain have changed significantly. On the one hand, telecom operators can provide users with more diversified services, a variety of new services are emerging. On the other hand, users can easily select different service models according to their preferences and actual requirements.The diversified services provided by telecom operators and the rapidly increasing number of users have generated massive data. Most of the data are records of behavior about users, which are of great value. How to make good use of these data is a new challenge for telecom operators.The paper focuses on how to use some machine learning techniques to analyze user behavior. This paper proposes a new clustering validity index based on frequent pair pattern. The new proposed index utilizes logical reasoning rather than compactness and separation measures that traditional indices usually use. The new index has overcome some inherent drawbacks of traditional clustering validity indices. In this paper, the new index and specific clustering algorithm are combined to get a clustering algorithm to automatically choose its optimal initialization parameter.In this paper, theoretical simulation experiments have been carried out to estimate the performance of the proposed clustering validity with artificial dataset and UCI dataset. The results demonstrate the new index behaves well compared with traditional indices. Then I have applied the new index to analyze real telecom user behavior with the dataset provided by China Unicom. The results show that the proposed index and corresponding self-adaptive clustering method are effective for practical application in the field of telecom user behavior analysis.