Agent-oriented Architecture for Task-based Information Search System

Lora Aroyo1 and Paul de Bra2
1Faculty of Educational Science and Technology, University of Twente, P.O. Box 217, 7500 AE Enschede, The Netherlands
aroyo@edte.utwente.nl

2Computing Science Department, Eindhoven University of Technology, P.O.Box 513, 5600 MB Eindhoven, The Netherlands
debra@win.tue.nl



Abstract

The topic of the reported research discusses an agent-oriented architecture of an educational information search system AIMS - a task-based learner support  system. It is implemented within the context of ‘Courseware Engineering’ on-line course at the Faculty of Educational Science and Technology, University of Twente, The Netherlands. AIMS is endowed with an explicit conceptual mapping of the domain knowledge structures and the user model in order to provide user effective educational information support. It is focused on task-based search, supported by conceptual information structuring and visualisation and user modelling. The overall information retrieval process is split into several sub-processes that are distributed among team of information agents. The agents’ behaviour is modelled according to the main activities involved in the process of task-based information retrieval. The system we describe is a knowledge-based course information support system whose main goal is to provide students and teachers with an intelligent help in respect to their preliminary defined curriculum tasks and goals. This way they are facilitated to handle efficiently the information within a specific subject domain. The students can use this system to perform and refine search queries for documents and to explore and navigate within the course domain structure. The teacher is facilitated with authoring environment for editing and construction of domain knowledge, course structure and documents database. We will present the general system architecture of AIMS and will give a detailed description of all system modules.

Pilot experiments are being performed within the 'Graphical User Interface Development' course in Eindhoven Technical University, to evaluate and validate the theoretical assumptions behind the AIMS visualisation and related functionality.

1. Introduction

Although the process of searching and finding relevant information is quite an old one, nowadays  it becomes very prominent within the context of the on-line and web-based education. Information retrieval, search results and information presentation are the main variables in this process. Their effectiveness is evaluated in respect to search hit rate, document access, search query formulation/reformulation, relevance determination, human-computer interaction and visualisation.

In this paper we discuss an agent-based information system that aims at providing combined adaptive information support for students and instructors within the context of on-line course environments. The main goal is to improve the usability and maintenance of information in such environments. AIMS supports the user in accessing, selecting and understanding the requested information. The idea is to integrate several technologies that supplement to each other in respect to more effective information retrieval within the framework of educational environments. The technologies we combine are intelligent agents, task-based search, concept map for information structuring and information visualisation, and user modelling as an overlay of the domain model (Aroyo, Dicheva, 99). Task-based search approach is considered to achieve better results in respect to the relevancy of the search results. Agent-oriented architecture provides quick and simple internal system communication, as well as flexible option for updating the system with new modules. Concept mapping technique allows for more flexible and dynamic knowledge representation, when supporting domain ontology. Concept map is an attractive information visualisation scheme in a human-like memory representation model (Collins, Quillian 70).

We discuss a prototype of an Agent-based Information Management System (AIMS) that implements the proposed approach. The reported work is performed within the framework of a Ph.D. research project related to the use of Agents technologies in Virtual Study Environments. AIMS was created to support the students from the Faculty of Educational Science and Technology participating in the course of ‘Courseware engineering’. The intention was to integrate AIMS in the web-based course environment already created for this course. We believe that the combination of agent technologies as system architecture together with information visualisation techniques in the context of concept mapping, task-based search algorithm and user modelling has a positive impact on the effectiveness of information search and better system adaptiveness in this process.

The paper starts with an introduction to the problems of educational information search and their solution proposal within AIMS prototype implementation. In the following section we will focus on description of the AIMS general architecture. The last part presents some conclusions and future perspectives.
 

2. AIMS General Architecture

Fig. 1 General System Architecture of AIMS


 


AIMS introduces the notion of agents as basic system entities, where agents are defined as self- contained problem solving software entities (Wooldridge and Jennings, 1995), which are autonomous, goal-driven and environment sensitive objects, situated within the environment and being able to sense and react to it, over time, in pursuit of its own agenda (Franklin, Graesser, 1997). Thus, an agent-based architectural framework is applied employing multi-agent team for task-based learner support. It involves also attractive visual representation of the information domain, and a conceptual organisation of information resources involving semantic mapping techniques (Aroyo, De Diana, Dicheva 98).

AIMS is built as an agent-oriented architecture. It includes the following main modules: Interaction Module, Query Module, Result Module and User Agent. As the system architecture is agent-oriented, the main activities within AIMS are carried out by agents, who are generally responsible for information retrieval and information presentation activities, domain knowledge representation and building, co-ordination activities among system agents, user profiling and system adaptation. As mediators between the users and the information content, agents are contributing to the general system adaptiveness. As a result of their collaborative intelligent behaviour, the system provides an intelligent user-oriented information support for learners and instructors.

The AIMS team of agents includes the following agent types:

2.1 Interaction Module

The Interaction Module is responsible for handling user input information from the two interaction modes:

The interaction module transforms the user input from those two modes in the form of terms, weights and documents and sends them to the other system modules via the user agent. User agent has strategies implemented to recognise the information coming from the two different modes and to interpret it for the other modules.

The components in the Interaction module are communicating with other system modules and entities via the Information Processor. It takes care for the integration of the results from Concept Map, Document Set and Search GUIs in a form readable for the other system entities. Concept Map GUI presents a view of the current request for terms and links in a form of a concept map. Document Set GUI presents the current result set of documents together with their descriptions and details. Search GUI presents the last request for user search query.

2.2 Query Module

The Query Module is responsible for processing and automatic reformulation of user's search query. Its components can be internal or external entities for the system. New components could be integrated in case of need for additional functionality. For the current AIMS purposes we have Synonym and Task agent, which are responsible of reshaping of user search query and adapting it to user preferences and tasks performed. Query Composer is responsible for the generation of a common query as an input for the Search Engine. It communicates with the search engine via the User Agent, by performing several iteration for query refine and final formulation. The User Agent is responsible for supplying the query to the search engine.

2.3 Result Module

The Result Module is responsible for retrieving and integrating the search results from the user's request. It performs searches within the domain knowledge base (documents, terms and links repository). It also takes care of results rating and refining with the user preferences implemented within the User Agent. Result module updates the status of the course assignments and topics progress. It works with Concept Map and Document Agents, which are generating set of result terms and documents for the users query. This module can work also with other external and internal agents, whose tasks are domain knowledge related. Course structure is also maintained within this module. The result module communicates with other system entities via the User Agent. It receives its input from the Result Composer, who is responsible for the integration of the result set from all the agents in this module

2.4 User Agent

User Agent is the central communication entity in AIMS system. User Agent contains strategies for User Model maintenance, information collection, and processing of user information from the GUIs to the agent pools. It receives the integrated user input from the Information Processor and distributes it respectively to Query and Result Modules. It also realises the communication of the system with the External World.

3. Conclusions and future perspectives

The topic of the reported research is the study of agent-based information management in dynamic and complex information environments, where the focus is on effective search in and use of complex information domains.

This paper presents an approach to effective and efficient information learner support by connecting together in one system intelligent agents, information visualisation and information classification techniques. We discussed an agent-based information system AIMS, where the core is a knowledge base containing the domain model, the user model and a set of rules for concept analysis and decision making. Concept mapping technique is used as a main knowledge structuring approach. This model allows organising the different topics around the key concepts (terms) of the domain. The Concept Map is also applied for the information visualisation and user presentation. The system capitalises on the advantages of visualisation approaches, agent technologies, information concept mapping and user-oriented task-based search approaches, while combining their strong points in instructional design context. It keeps the balance between visual and textual information presentation, detailed and general information views, task-based and traditional search approach, user-centred and fully controlled educational environments.

A pilot experiment is being performed to evaluate and validate the theoretical assumptions behind the AIMS visualisation. Sixty students from the ‘Graphical User Interface Development’ course in Eindhoven Technical University, the Netherlands, are taking part in the experiment. All students were split randomly into two groups - control and experimental. The students in both groups were introduced to the AIMS user interface. The system, which the students in the control group are using is with disabled ‘task-based search’ option. The experimental group is allowed and encouraged to use the ‘task-based search’ option. Within these major groups the students were also divided into smaller groups of three in order to complete a course assignment. The students were given the http link to AIMS, an experimental assignment, a questionnaire and a time checking table. The time checking table was handed to them in order to take measures of time in previously assigned breaking points. The course assignment includes course tasks related to user interface evaluation methodologies and the experimental assignment conducts the steps and order of the experiment. The experimental study is conducted in the period of  15 days and the size of the groups remains constant through the duration of the study. As a measurement instrument for this experiment a questionnaire is used constructed especially for this study. The time checking table and the student assignments are also planned as measurement instruments.

The results of the current experiment will be used as input for a second pilot experiment, which will take place in November 1999 with students from the Antwerpen University. Some of the suggestions and recommendations of the students involved in the first experiment will be used for improving the AIMS GUI before the second experiment. The results of both experiments will be summarised and will serve as a basis for developing a new improved version of the system.
 
 

References

Aroyo, L., De Diana, I., Dicheva, D. (1998). Agents to Make Your Information Meaningful and Visible: An Agent-Based Visual Information Management System, Proc. of WebNet'98, Conference, AACE, Orlando, USA.

Aroyo, L. & Dicheva, D. (1999). Information Retrieval and Visualisation within the Context of an Agent-based Information Management System.Agents that Make Your Information Meaningful and Visible: An Agent-Based Visual Information Management System, Proc. of the EdMedia’99 Conference, AACE, Seattle, USA.

Collins, Quillian, . (1970) Semantic Memory. Readings in Cognitive Science. California: Morgan Kaufman Publishers.

Franklin, S. & Graesser, A. (1997). Is it an Agent, or Just a Program? ECAI Workshop, 1996, Hungary.

Travers, M. (1996). Programming with Agents: New Metaphors for  Thinking About Computation, Massachusetts Institute of Technology. Available through WWW: http://mt.www.media.mit.edu/people/mt/thesis/mt-thesis.html.

Wooldridge, M. & Jennings, N. (1995). Intelligent Agents: Theory and Practice, Knowledge Engineering Preview, 10(2).