At present, with the increasing personalized demand of campus network users, the use of network information service has changed from a centralized and unified type into personalized distribution. Campus network users have their surfing habits and interests. This makes differences between the surfing time, internet service and the focus point. Campus network services also changed from the traditional network chatting, reading news on page and other aspects into a more detailed and more personalized services, such as shopping online,classroom online, game online and so on. That provides great convenience for campus network users. However, the individual campus network user research in the timeliness,stability and other aspects cannot meet the growing demands. Thus, provide more timely and higher quality network services is a trend, for the research of network user behavior. This paper will use the data mining, clustering analysis technology to study the behavior of network users, and make a further analysis of the needs and interests of network users, in order to optimize the user experience. The network users of campus in colleges and universities have relatively high educational level and professional level, and different users in campus network has a different surfing habits, and more personalized network demand. Therefore,the campus network data is more researchable compared social network data.The main research work and innovations of this paper are summarized as follows:1.This paper reviews the current situation of network user behavior analysis in recent years at home and abroad about. Summarize the concept and characteristics of campus network user behavior, the process and methods of data mining. This paper selects the improved K-means clustering algorithm, based on the characteristics of the campus network user behavior analysis of the data. Then the algorithm can be used for cluster analysis.2.This paper proposes a similarity computing method based on the interest degree matrix based on the traditional K-means clustering algorithm. This algorithm improves user similarity calculation method through constructing a user interest degree matrix, that is the user’s behavior online is expressed in the form of a matrix. Each matrix has 7 rows which respect days include Monday to Sunday, and each matrix element shows the interest proportion of the same day under a constraint, and each column shows the interest proportion in a last week, and then calculate the similarity with clustering technique.In a certain extent, this algorithm reduces the invalid data, and improves the accuracy and effectiveness of clustering user behavior similarity. It also can be better applied to the concrete research of campus network user behavior.3.In this paper, we propose an algorithm for screening of the conditions that affecting the user’s behavior. That is, according to the user’s behavior with the conditions of fluctuations to make the effective determination of the conditions. There are many conditions affecting the users’ behaviors in the local area network, such as surfing time in a day, surfing place and soon. Thus, scientifically find out the effective conditions is an important researching part of the campus network user behavior. Therefore, Compared to the traditional manual screening method, this algorithm reduces the subjective randomness of human intervention and the results are more rigorous. |