next up previous
Next: Summary Up: Task Planning Agents in Previous: TPA knowledge

Implementation

There is a working prototype of the UMDL and of a TPA. The current UMDL contains several hundred collections. There is a BSOA that uses a locally developed subject hierarchy specializing in earth and space sciences, and a TA that uses data from the NASA thesaurus. The implementation of the TPA basically follows the description given in this paper except for the economic or deal-making aspects of it, which have not yet been implemented. The TPA is capable of receiving a query from the UIA, consulting the RA, BSOA, and TA, and returning a set of appropriate collections based on the information it gathered and on its local knowledge. It can also talk to several UIAs at the same time, interleaving the different conversations and handling messages as they arrive. The current working prototype can be accessed at [2].

This implementation served to verify the effectiveness of our agent architecture and the use of UM-PRS as a basic platform for developing software agents. As we described before, the query answering plans are rather simple, and the semantic knowledge sources (i.e. the BSOA and TA) are also simplistic, of narrow scope and limited in number. Also, the collection topics are currently registered using only a limited subset of the BSO language. All this implies that the search carried out by the TPA is slower and less accurate than we would want. We are currently working on increasing the TPA's search knowledge, while other members of the UMDL are working on improving and adding more agents like the BSOA and TA, and on expanding the descriptions given to the collections. Given these improvements, we hope that our next version will be faster and much more tailored to the specific user's needs. We also plan to include some basic pricing/buying strategies and more complex contracting/team-forming capabilities into the TPA and other agents. These changes should start to move the UMDL towards becoming a real economic society.


next up previous
Next: Summary Up: Task Planning Agents in Previous: TPA knowledge

Jose M. Vidal
jmvidal@umich.edu
Mon Mar 11 15:41:27 EST 1996