KOINOTITES is evaluating different machine learning methods on the task of constructing communities of users in digital libraries.
The separation of “wheat from hay” on the Web is becoming increasingly important as the amount of information available electronically becomes unmanageable for its non-expert receivers. The amount of information provided by a digital library (its breadth and coverage) is becoming less significant than the ability to access the required information at the right time and in the most suitable form. The personalization of Web services and in particular digital libraries is a leap in the direction of alleviating the information overload problem and making the Web a friendlier environment for its users.
KOINOTITES (transliteration of the Greek word for Communities) is a project funded internally by NCSR “Demokritos” to study the effect of personalization in digital libraries. In the course of the project, unsupervised machine learning methods are evaluated on the task of constructing models for communities of users with common interests and/or common navigational behaviour in the site of a digital library.
KOINOTITES combines methods from two very active research fields: User Modelling and Machine Learning. The result contributes to the development of both fields.
Adaptive modelling of the users of a digital library. Machine learning methods promise to make user modelling applicable to dynamic environments, such as online digital libraries. The automatic construction of user communities is one of the most promising paths in this direction.
Critical comparison of learning methods for large-scale analysis of Web usage data. Web usage data are particular in many respects, e.g. the sequential nature of navigation and the large scale of the data. Our study focuses on the suitability of different data mining methods for these data.
The library of NCSR "Demokritos" and the Web site of the Inst. of Physical Chemistry (NCSR "Demokritos") are used in the evaluation of the methods. The Web site of the Inst. of Physical Chemistry provides a public information service for researchers in Chemistry. This service is called "Information Retrieval in Chemistry" and includes a large number of links to related information sources.