Research On Technique Of Information Filtering Based On CoP Modeling

Today, more and more enterprises adopt Information Management or Knowledge Management Systems to enhance efficiency. Actually, the employees in such systems are working in a virtual collaborative environment. They need timely and valid information support related to their tasks. Traditional information filtering techniques, which take the interests of users into account only, fail to satisfy this requirement. In this thesis, we introduce an information filtering approach based on CoP (Communities of Practice) modeling, and investigate its key techniques. The overall research effort has been broken down to a set of detailed research works:Aiming at the users' requirements for information in enterprises col laborative environment, this research offers an information filtering approach based on CoP modeling. Knterprise employees in a collaborative environment often face new jobs or tasks. Because they are unfamiliar with the new tasks, they cannot provide related information requirements. Thus, traditional information filtering techniques are unable to produce timely and valid informat ion support to these employees. CoP means a kind of teams that are set up u> share knowledge and help team members to learn from each other during work. Its profile is the reflection of the tasks of team members. Based on former information filtering approaches and the I)-S theory, this thesis gains CoP' s profile by model ing the interest of CoP, studies and implements the CoP orientated information filtering technique. A domain-based vector space model is provided by this research. The denotation models of information, user interests and CoP interests in an information filtering system should be consistent. The most popular vector space model is direct, concise and easy to be implemented. However, it can only express the user-interested keywords and cannot distinguish the different interests of users well. Furthermore, the large amounts of keywords result in the Jail of algorithm' s efficiency. To solve this probiem, this thesis prov i des a domain- based vector space mode 1 by bu i 1 di ng a classification model and offer i ng the method to cal culate the documents' weight in the domains. This model can greatly reduce the number of dimensions, incarnate the variety of users' interests wel 1, and adopt many mature techniques in the vector space model such as comparability formula.Based on some traditional algorithms, this thesis provides a weighted star cluster algorithm to learn user profile. The profile of CoP is fused from the profiles of its members, so the gaining of the user profiles is the base of CoP modeling. The learning algorithms to build user profile are the hot point in the current researches of information filtering. Cluster algorithm integrates the advantages of Rocchio and kNN. But conventional cluster algorithms cannot embody the different interests of users. This thesis brings forward a weighted star cluster algorithm that can reflect the users' information requirements better through focusing on the documents in which users are most interested to build clusters.Our investigation provides a valuable reference for the future work of information filtering in collaborative environment.