Personalized Service Based On Web Log Mining Research And Applications

With the rapid development and continued expansion of Internet, the explosive growth of web information has caused"information overload", which is becoming a serious obstacle for people to get information they need. How to provide personalized service quickly and accurately becomes a great challenge. The personalization service improves acquisition efficiency of web information and meets the individual needs of users, having important theoretical meaning and practical value. Personalized service based on Web is becoming a hotspot in web mining and core of personalized service systems.Aimed at the main challenges of web mining and personalization, this thesis focuses on some key technologies of personalized recommender systems, such as data preprocessing in web log mining, user interest model and recommendation algorithm, etc. The main contents and achievements of this dissertation are as follows:1. Analyzes the importance of data preprocessing, describes the data sources and format, enumerates stages of it in details, and finally gives a heuristic algorithm.2. The user interest model representation and discovery technology are outlined, with their respective principles, features and application. Specially dissect the K-Means algorithms for clustering, including the process and limitations. So we use an adaptive density-based K-Means Algorithm. Experiment results indicate that the new algorithm improves the quality of clustering significantly.3. This thesis classifies the personalized recommendation technology, analyzes the characteristic, and points out the shortage of information theory based algorithms. Integration of information theory and user clustering algorithm is proposed. Experiments prove that the new algorithm improves recommendation effectiveness and precision at relative low performance loss.4. Design and implement a web log based personalized prototype recommendation system. A hybrid recommendation framework is used for the unregistered and new users'problem, and improves the accuracy of recommendations for registered users.