Free Text User Request Processing in the System “KSNet” Alexander V. Smirnov, Mikhail P. Pashkin, Nikolai G. Chilov, Tatiana V. Levashova, and Andrew A. Krizhanovsky St. Petersburg Institute for Informatics and Automation, 39, 14 th line, St. Petersburg, Russia {smir, michael, nick, oleg, aka}@mail.iias.spb.su Abstract This paper considers implementation issues of multi- agent systems dealt with knowledge management; it describes the implementation of problem-oriented agents designed as a part of the being developed KSNet- approach. This approach addresses the problem of knowledge logistics and considers it as a problem of knowledge source network configuring. Utilizing intelligent agents was motivated by a distributed and scalable nature of the problem. This paper describes an implementation of translation agent (user request parsing). 1. Introduction Comfortable user’s work (and pleasure from high- quality results) depends on used programs. Usability is a qualitative system parameter consisting of habits (with ease, clarity, and accessibility), design simplicity, and functional completeness [1]. One way to overcome these contradictory requirements is to formulate user requests in natural language. The restrictions of implementation (but not of approach) are that the request should be worded in English for one of several predefined problem’s domains. The possibility to formulate user requests in free form (sec. 3), use of ontologies (sec. 2), and multi-agent systems (MAS) [4] are three remarkable features of the considered project. MAS is the promising branch of computer science. But effective MAS using requires that a number of problems to be solved, such as: the information discovery problem, the ontology problem, the legacy software problem, the reasoning problem, and the monitoring problem [2]. Some of these tasks are worked out by knowledge fusion (KF) methodology. The knowledge fusion (KF) methodology is a new direction of knowledge management related to knowledge logistics [3]. User-oriented MAS would need to continually model its users. Such user modeling would be vital for any personalization of the assistance it provides to its user. The “user profile” paradigm is used in the developed system. It consists of both explicit (user enters data himself) and tacit information (parsed requests) about users. However there are outstanding several challenges related to profile [2]: • How to detect interest migration and how to maintain an up-to-date profile? • How to test (qualitatively) its usefulness? • Work towards profile standardization is needed. The paper is organized as follows. Section 2 elucidates the knowledge logistics concerned with the ontology approach. Section 3 presents the translation agent design, and its implementation. Main features of the translation agent in the described system are summarized in the conclusion. 2. Ontology-Driven Knowledge Logistics Knowledge logistics addresses the problem of acquisition of the right knowledge from distributed sources, its integration and transfer to the right person within the right context, at the right time, for the right purpose. This problem in the approach is considered as a configuration of network that includes end-users, loosely coupled knowledge sources, and a set of tools and methods for knowledge processing located in e- business environment. Such network of loosely coupled sources was referred to as knowledge source network or “KSNet”. The application of intelligent agents representing their knowledge via ontologies [5] was motivated by the need of knowledge logistics systems for flexibility, scalability, and customizability. The multiagent system architecture based on the FIPA Reference Model [6] was chosen as a technological basis for the definition of agents’ properties and functions since it provides standards for heterogeneous interacting agents and agent-based systems, and specifies ontologies and negotiation protocols. The formalism of object-oriented constraint networks [3] was chosen as the abstract model for ontology representation. According to the formalism knowledge can be described by classes, attributes, domains, constraints, and methods. This perspective of knowledge representation correlates well with the semantic metadata representation concept being developed under the Semantic Web project [7]. The thorough comparison of multi-agent systems (KRAFT, InfoSleuth) designed for knowledge fusion operations vs. the system "KSNet" has been done in previous works [4]. The system "KSNet" has a distributed multiagent architecture [4]. Components of the system can be allocated at different hosts and connected via TCP/IP protocol. Users and experts can work with the system via a Web-based interface. SPECOM’2004: 9 th Conference Speech and Computer St. Petersburg, Russia September 20-22, 2004 ISCA Archive http://www.isca-speech.org/archive