Multi-Agent Visitor Advising - MAVA

Paul Lockier & Pete Edwards

The World Wide Web provides the business traveller/tourist with a vast online repository of information on destination cities, accommodation, entertainment, flights, etc. Users wishing to interact with such Web-based information resources typically encounter standard form-based query interfaces. We are investigating how such interfaces can be augmented with intelligent agents which reason about the likely success of a user query, and if necessary, adapt the query (through the use of background knowledge) to increase its chance of success.

MAVA (Multi-Agent Visitor Advising) employs a number of such agents to mediate access to a collection of databases containing information on hotels, restaurants and pubs in the Aberdeen area. The Bellboy, Maitre d' and Landlord agents each monitor queries submitted via a standard HTML form interface to their respective databases, and if the number of hits generated by a query is below a threshold value, attempt to adapt the query. Each agent employs its own specialist knowledge base which contains details of the acceptable modifications to user queries. For example, Bellboy is able to relax the requirement that a hotel room have a four-poster bed (if this is resulting in few matches within the hotel database), but is unable to alter a request for a double room into a request for a single. Agents are also able to explain why they recommended a certain hotel/restaurant/pub.

In addition to adapting user queries, each of the MAVA agents is also able to observe user activity and share meta-level information with the other agents. For example, if the Bellboy observes that a certain user is searching for expensive hotels in a particular geographical area (say, near the airport) it can communicate this to the Maitre d' and Landlord agents. Should the user subsequently query either of the other databases during the same session, these agents can make use of the information as additional knowledge to guide the query adaptation process.

Bellboy	Maitre'd	Landlord

Click on the Bellboy & Landlord icons above for illustrative screendumps. WARNING: Images may take some time to download.

Click here to visit MAVA.

Publications:

• Exploiting Learning Technologies for World Wide Web Agents
  IEE Colloquium on Intelligent World Wide Web Agents, Digest No: 97/118, IEE, Savoy Place, London, 3/1 - 3/7, 1997
  P Edwards, C L Green, P C Lockier & T C Lukins

Funding:

Internal Undergraduate Project.