PROJECTS IN ISL

Most projects in ISL have topics related to a flexible/intelligent dialogue in query-answering systems. A query-answering system is a system that manages a large base of information in such a way that users can pose queries to and get answers from the system about the information stored. Examples of query-answering systems range from dedicated query applications (electronic web-services, for instance) to general database systems. A query-answering system often replaces a human with expertise in the domain of the information base. "Intelligence" in this context concerns the capability of the system to mirror some of the human experts' qualities regarding preferred behavior in a query-answering dialogue. This includes attempts to interpret also incorrectly posed questions and to compose answers reflecting also knowledge about the user and the domain. Also inclusion of related, not strictly queried, information in answers, when relevant, is good "answering"-behaviour and should therefore be expected from the "intelligent" system. Most projects in ISL are part of a program called "Intelligent Search-Systems" supported by the Danish National Science Foundation. Below a brief description of selected ISL projects. Find more at http://www.isl.ruc.dk.

Discovery and use of associations
In this project we work with an approach to database querying based on fuzzy evaluation of queries, where similarity knowledge hidden in the database will be exploited in query answers. A dedicated mining for domain knowledge is involved, the result of which is a special kind of associations between values. These associations are represented in fuzzy networks, that in turn are used to relax queries and to relate objects in the database. An implementation of one direction within this approach is used in the DanBib project.

Flexible Search in DanBib
Since spring 1996 ISL has, in this project, collaborated with the company DBC Ltd. (Danish Library Centre Ltd.) - the provider of the system DanBib, that comprises a union catalogue for all libraries in Denmark. The purpose of this project is to analyse the potentials in, and to do experiments with a range of possible improvements of techniques for search in bibliographic databases, with special focus on the DanBib system. A prototype system has been developed, demonstrating the potentials of fuzzy evaluation techniques in general, and discovery and application of a fuzzy semantic network, in particular. The prototype is a www-application and is available in several versions (including a pure Danish and a pure English) at http://www.isl.ruc.dk/proto_index.html

User profiling
In a user profile the system represents knowledge about the user to be applied in customizing the dialogue with the user. At present we work on an approach to automatically collect a user profile from the user's behaviour in the query-answering dialogue. Knowledge about topics of interest and "preferred" domain values are examples of what collected to the user profile.

Query relaxation from taxonimies
In this project we apply meta-programming in logic to perform query relaxation based on taxonomies (or hierarchical thesauri). The idea is that if a query has empty answer and a constraint in the query has a more general counterpart in a taxonomy, e.g., a travel as more general than a flight, then the query can be relaxed by replacing with the more general constraint.

Hypothetical queries
Hypothetical queries are queries embedding hypotheses about the database. In principle, the answer to a hypothetical query h>q with a hypothesis h is the result of evaluating q against the database updated with h. In the project we have developed an approach to negative hypothesis: "counterfactual exceptions", which as compared to the other approaches can be efficiently evaluated and have a simplified semantics,. A prototype system implementing counterfactual exceptions is under development.

Fuzzy Query Weakening
We work here with an approach to the problem of empty answers to conjunctive fuzzy relational queries. We do experiments with two weakening mechanisms generating more tolerant queries, that may be used partly in obtaining non empty answers and partly as a basis for explaining query conflicts.

Dynamic fuzzy terms
We focus in this project on application of special purpose fuzzy terms, in access to databases, that dynamically defined in the sense that their definitions vary with content of the base.. Dynamic terms may reflect the current state of a database only, or some recent history - e.g. the last 3 months. Dynamic terms are supposed to be used in queries. Examples are "reliable" customer, "popular" book, "good" salesman, an "average" student.

BookHouse on WWW
In collaboration with the Royal School of Librarianship, Denmark, ISL is working on a project involving further development of the BookHouse system a visual library system (originally developed at RIS under leadership by Annelise Mark Pejtersen). A first goal in this project is a WWW-application, based on the ideas of the BookHouse system and including fuzzy query evaluation. A prototype is at present available in Danish at http://www.isl.ruc.dk/boghus