A Virtual Community Library: SICS…

                    A Virtual Community Library:
              SICS Digital Library Infrastructure Project

                    Andreas Rasmusson, Tomas Olsson and Preben Hansen

         Swedish Institute of Computer Science (SICS), Box 1263, SE-164 29 Kista, Sweden
                                    {ara, tol, preben}@sics.se



1 Introduction

In this project1, we aim to create an agent-based digital library architecture for a
Virtual Community Library (VCL) where each user has a personal library and, at the
same time, is part of a larger community. The community is dynamically composed
of the users' personal libraries and, through intermediators, other digital libraries.
   We want to stress the fact that the users participate in a large dynamic de-
centralised community where they continually interact with each other. Being a part
of a community means that each user can benefit from the work put into the other
libraries. For example, by obtaining documents through search queries or recom-
mendations using social filtering, but also by getting help to organise the personal
library.
   In the VCL, we try to combine the best aspects of the WWW, the library and the
personal library. For example, ease to publish documents, personal information
space, decentralised control of the document collection and ability to search for
documents.
   We have currently implemented two prototypes of the system, one for the per-
sonal library and one for visualising the information spread between the users.


2 Agent Architecture

The foundation for the VCL is an agent architecture where the users are represented
by self-interested agents. This is an open-ended knowledge system, which supports
creation, inferring, manipulation and sharing of knowledge about information ob-
jects ("metadata"), and supports (enables automatisation of) interaction between
agents pertaining to the information and knowledge management related business
processes.
   Interaction (compatibility) with other systems in a number of formats and proto-
cols will be investigated, such as Z39.50, MARC, DIENST, Dublin Core, BibTeX,
etc. [1], [2]. Many existing bibliographic formats are ambiguous or limited in the


1
    See http://www.sics.se/isl/diglib
way they represent knowledge. This introduces difficulties when the information is
to be used in machine-machine conversations (in contrast to human-machine). In our
architecture, emphasis is put on handling translations between partially incomplete,
incoherent or incrementally developed ontologies and particular collections of
knowledge.
   The work is to be based on software-agent research and platforms developed in
the SICS Intelligent Systems Laboratory.


3 Personal Library

A user is represented by a personal library agent. By emphasising that the informa-
tion put into it is relevant to the individual users we address the problem that users
will rarely bother to register information they themselves have no interest in. Classi-
fication of information by end-users usually introduces the problem of having non-
librarians doing classification according to the best of their own knowledge. This
makes it harder to develop usage-conventions in classifications. Hence, we still need
librarians that classify larger collections to get homogeneous usage-conventions.
   Within the information architecture, no library has any special status as keeping
complete, correct or authoritative information. Each digital library is autonomous
and free to negotiate which other libraries it depends on to find new information.
However, large cohesive collections (institutions) should still be "influential" in
defining authoritative descriptions of objects since they are more often consulted for
information than a single-user personal library.


4 Information Dissemination and Retrieval

By seeing our digital library not just as a static entity but as a dynamic virtual com-
munity we strive to get properties of real communities. In real communities, infor-
mation is spread in a number of ways, such as by looking in somebody's personal
library, asking or telling other persons about something or searching in a database.
The control of the information flow is decentralised to the individuals of the com-
munity and they can retrieve information both actively (by asking or looking) and
passively (by being told). We implement active retrieval as explicitly formed search
queries and we implement passive retrieval as decentralised social filtering based on
trust [3], which is built on the interactions between personal recommender agents
performing content-based filtering. Hence, we try to combine the advantages of both
filtering techniques.
References

1. Library of Congress: Metadata, Dublin Core and USMARC: A Review of Current Efforts.
   MARBI Discussion Paper no. 99, Library of Congress, January 21 (1997)
2. Heery, R.: Review of Metadata Formats. In: Program, Vol. 30, No. 4 (1996) 345-373
3. Olsson, T.: Decentralised Social Filtering based on Trust. In: The Working Notes of the
   AAAI-98 Recommender Systems Workshop, Madison, Wisconsin (1998)