Research and Implementation of Personalized Recommendation Services by Data Mining Techniques

¡@

As the continual increase of the population on the Web and the constant improvement of web sites, how to emerge from thousands of web sites and attract the users has become a very hot and important issue. The key to win such a tournament is to capture the user¡¦s need and then provide the right services for individual users. Reasoning based on the past behaviors is a good way to capture the user¡¦s need.

We will apply and refine the data mining techniques to derive user¡¦s information needs and then provide the suitable information for recommendation. This project will proceed in three stages as follows:

(1) Applying the traditional data mining techniques to personalized recommendation

In the first stage, we adopt the traditional way of association rule discovery to derive user interests and behaviors for browsing from the log data. Based on the derived profiles for individual users, we further classify the users into clusters. At last, the clustered results are used for personalized recommendation services. In this stage, we will design and implement a prototype for thesis recommendation, called the online thesis system (OTS), to testify the effectiveness of the derived profiles.

(2) Research of the multiple-sort data mining techniques

In the second stage, we further consider the multiple-sort property. A data structure called the multiple-sort hierarchical tree will be built and used to derive the hierarchical association rules. In this stage, we will analyze the performance of the related works on data mining and compare their efficiency with our method by some simulations.

 (3) Applying the multiple-sort data mining techniques to personalized recommendation

In the third stage, we apply the multiple-sort property to profile derivation, including the mining of periodic user interests and behaviors, and the ones for multiple-sorts. According to different mining techniques, we will provide the corresponding recommendation services, respectively. Finally, we will compare the quality of these services to find out the influence of different mining techniques upon a personalized system.