| Abstract |
The aim of this thesis is the development of a system that provides recommendation services, which are adapted to the user's behavior and are based on the dynamic management of communities of network users. The term recommendation system refers to a system that has the ability to collect and introduce to each user documents (in the broad meaning of the term), which belong to the field of her interests. In the recent years an increasing interest in recommendation systems has emerged both from the research and the application point of view, and in both academic and commercial/economic domain. Modern recommendation systems mainly base their functionality on two ways, in order to suggest/recommend a document to the user. These two ways are content-based and collaborative-filtering. If these two functional ways are applied separately, they present crucial disadvantages and suffer from specific problems. In order for these problems to be dealt with, hybrid approaches are adopted, where techniques that are applied in both types of recommendation systems are used. In the context of this work, a general recommendation algorithm, which adopts a hybrid approach, is suggested. This algorithm is based on the maintenance of a user-profile for each user and on the dynamic adjustment of this profile to the users behavior. Moreover, this algorithm is based on the dynamic management of communities, which contain "similar" and "relevant" users. These communities are created according to a suggested algorithm for classification of "relevant" users into communities. This algorithm is implemented by a community management mechanism that has been developed, in the context of this work. The majority of the comparison and relevance techniques are based on set-operations or on application of vectors internal product. In both cases however, the "identical-ness" of terms and not the real relevance between them is examined. The comparison mechanism that has been developed and is used, in the context of this work, is based on semantic relevance between terms. In order for this semantic relevance between terms to be feasible, a glossary of terms is used. As a further aim the application of this algorithm for classification of the users in communities on one hand, and the practical hypostasis of the proposed recommendation algorithm on the other, we have expanded the systems services. In this way, the ability of a user to register and thereafter login into the system is provided. Moreover, the capability of searching in digital libraries, which make their data available via Z39.50 servers, is provided. At the same time, the service of handling documents is available, which is responsible for the documents storing and rating by the user. Furthermore, a mediating schema (Mediator), which provides the system with the capability of extension and the flexibility to dynamically accept new services, has been developed. Lastly, the system is provided with a user interface, which has been developed in a way that enables the accessibility of the system using applications for browsing the World Wide Web (Web Browser). All the data of the system are stored in a database. Aiming at the independence of the system from each occasional database system, an Application Interface (API) for various database management systems has been developed. This API enables the communication of the system with a range of database management systems.
|
Ανάπτυξη Συστήματος Προσαρμοζόμενων Υπηρεσιών Συστάσεων βασισμένο στη δυναμική Διαχείριση Κοινοτήτων Χρηστών
