Refinement of the ACP2P by sharing user-feedbacks and learning query-responder-agent-relationships

ABSTRACT

This paper proposes two methods for improving the retrieval accuracy of the Agent-Community-based Peer-to-Peer information retrieval (ACP2P) method. One uses user feedbacks exchanged in a community. The other uses query-learning methods that make a middle agent to learn query-responder agent relationships. The latter methods are useful not only for improving the retrieval accuracy, but also for reducing communication loads. We conduct several experiments with test collections so that the evaluation can be done in an objective manner. The experimental results illustrate the validity of our proposed methods.

REFERENCES

• 1: T. 3rd and 4th NTCIR Workshop Data. http://research.nii.ac.jp/ntcir/{ntcir-ws3/work-en.html#data,ntcir-ws4/data-en.html}.
• 2: T. Mine, D. Matsuno, A. Kogo, and M. Amamiya. Design and implementation of agent community based peer-to-peer information retrieval method. In CIA2004, volume LNAI 3191, pages 31--46, 9 2004.
• 3: S. E. Robertson, S. Walker, S. Jones, M. M. Hancock-Beaulieu, and M. Gatford. Okapi at trec-3. In NIST Special Publication 500-226: TREC-3, 1994.

INDEX TERMS

Primary Classification:
  H. Information Systems
  H.3 INFORMATION STORAGE AND RETRIEVAL
      H.3.3 Information Search and Retrieval

General Terms:
Algorithms, Design, Experimentation, Performance

Keywords:
ACP2P, information retrieval, multi agent, peer-to-peer, query learning, user feedback