基于机器学习的电信网络用户行为分析研究

发布时间：2018-06-06 08:26

本文选题：聚类 + 聚类有效性　；参考：《北京邮电大学》2014年硕士论文

【摘要】：随着移动互联网时代的到来以及移动智能设备的普及,传统的电信业务模式和用户能够获得的服务内容发生了显著的变化。一方面,电信运营商能够给用户提供更为丰富的服务内容,各种新业务不断涌现；另一方面,用户可以根据自己的个人需求和喜好选择不同的服务内容和模式。电信运营商多元化的服务和用户数目的迅速增长产生了海量数据。这些海量数据中有大量的用户行为数据,这些数据具有巨大的价值,如何利用好这些海量的用户行为数据是电信运营商的一个新挑战。本文的主要创新点如下：首先,本文研究了如何利用机器学习中的相关技术针对电信业务用户行为数据进行用户行为的分析和挖掘,并且提出一种基于频繁模式的聚类有效性指数。新提出的聚类有效性指数不再采用传统聚类有效性指数经常利用的内聚性度量和分离性度量而是基于统计逻辑推理度量。它克服了传统聚类有效性指数的一些固有不足。文章将新提出的聚类有效性指数与特定聚类算法结合,形成了一种自适应聚类方法。这种自适应聚类方法能够从具有不同初始化参数的多个聚类过程中选择一个最合理的初始化参数和相应的聚类结果。其次,本文将提出的自适应聚类方法应用于人工数据集和UCI (UC Irvine)数据集进行理论仿真和分析。对比实验结果表明,在人工数据集和UCI数据集上,新提出的有效性指数具有较好的性能。然后,利用本文提出的自适应聚类方法对中国某运营商某省级公司提供给本实验室的移动业务用户行为数据实施了特定的自适应聚类分析任务,结果表明本文提出的新聚类有效性指数和自适应聚类方法能够有效地应用于真实的电信业务数据以分析用户行为,具有实际应用价值。
[Abstract]:With the advent of the mobile Internet era and the popularity of mobile intelligent devices, the traditional telecommunication service mode and the service content that users can obtain have changed significantly. On the one hand, telecom operators can provide users with more abundant service content and a variety of new services are emerging; on the other hand, users can choose different service content and mode according to their personal needs and preferences. The diversified services of telecom operators and the rapid growth of the number of users have produced huge amounts of data. There are a lot of user behavior data in these massive data, these data have great value, how to make good use of these massive user behavior data is a new challenge for telecom operators. The main innovations of this paper are as follows: Firstly, this paper studies how to analyze and mine the user behavior data based on the related technology of machine learning, and proposes a clustering validity index based on frequent pattern. The new clustering validity index is based on the statistical logic reasoning instead of using the cohesion and separation measures which are often used by the traditional clustering validity indices. It overcomes some inherent shortcomings of traditional clustering validity index. In this paper, an adaptive clustering method is developed by combining the new clustering validity index with a specific clustering algorithm. This adaptive clustering method can select one of the most reasonable initialization parameters and corresponding clustering results from multiple clustering processes with different initialization parameters. Secondly, the proposed adaptive clustering method is applied to the theoretical simulation and analysis of artificial data sets and UCI UC Irvine data sets. The experimental results show that the proposed validity index has better performance on artificial data sets and UCI datasets. Then, using the adaptive clustering method proposed in this paper, we implement a specific adaptive clustering analysis task for mobile user behavior data provided to our laboratory by a provincial company of a Chinese operator. The results show that the new clustering validity index and adaptive clustering method proposed in this paper can be effectively applied to real telecom service data to analyze user behavior.
【学位授予单位】：北京邮电大学
【学位级别】：硕士
【学位授予年份】：2014
【分类号】：TN929.5;TP393.01;TP181

【共引文献】