| Mining the pattern of visiting websites is one of the most important applications in present data mining. The user navigation pattern mining can make the websites builders understand clearly different interests of their users and the visiting situation of their whole web pages. On the other hand, in order to fit different types of user accesses, it can adjust endlessly the logic organizing structure of the web or set up self-adaptive website.Because web is unstructured and dynamic, it can't be mined directly. However, Web log file has integrated structure. So it is realized to mine Web log to customize service of website.In this dissertation, UNPMRS(User Navigation Pattern Mining Recommending System) was designed by researching previous framework of electronic commerce recommendation system. Then its every module's function and how they correspond and work together was expatiated. Data preprocessing and frequent patterns mining were focused. Data preprocess is a critical step in Web Usage Mining. The results of data preprocessing are relevant to the latter steps, such as path analysis, user navigation pattern clustering, association rules mining, and so forth. The thesis carried a research on the most popular method of data preprocess, then the dividing historical navigation sessions into frames of sessions based on day interval was presented according to the real need of the cluster algorithm in UNPMRS, as result of that, the efficiency of the data preprocess was enhanced, and got a sound trail conclusion.In this dissertation, a new recommendation method was proposed, which was applied to web log mining by integrating user clustering and association rule mining techniques to improve the effectiveness of electronic commerce recommendation. The experiment results showed that the method could improve the precision rate and F rate of evaluation criteria effectively. Experiments conducted on real Web server logs verify the usefulness and practicality of this technique. |
PDF Full Text Request