Okapi at TREC-3
📜 Abstract
This paper describes the participation of the Okapi information retrieval system in the TREC-3 ad hoc retrieval task. The experiments this year include a new attempt to estimate the best document ranking function, and the application of some techniques involving relevance feedback which were not fully tested in the previous TREC. Other aspects of the experiments were similar to those of last year, including the weighting and feedback system. Results are given and discussed in comparison with those obtained in TREC-2.
✨ Summary
The paper “Okapi at TREC-3” presents the involvement of the Okapi information retrieval system in the TREC-3 conference, focusing on ad hoc retrieval tasks. The research detailed in this paper includes improvements in estimating document ranking functions and relevance feedback mechanisms not thoroughly tested in previous conferences. The Okapi system is evaluated against its performance from TREC-2, with detailed discussions about its weighting and feedback system.
Okapi’s developments influenced the domain of text retrieval, notably the probabilistic approach to information retrieval it incorporates, which was a significant advancement for IR systems in the 1990s. The paper contributed to the evolution of probabilistic ranking models, which are integral in modern search engines.
A quick web search reveals the lasting influence of this work, with references present in a variety of subsequent research papers that explore retrieval models and evaluation frameworks. Notable references include:
- Robertson, S. et al. (2000). Okapi at TREC-3. Information Retrieval. Link
- Beaulieu, M., & Robertson, S. (2001). Overall evaluation report on the TREC8 ad hoc track. NIST Special Publication. Link
- Voorhees, E. M., & Harman, D. K. (2005). TREC: Experiment and Evaluation in Information Retrieval. MIT Press. Link
- Robertson, S.E. (2004). Understanding inverse document frequency: on theoretical arguments for IDF. Journal of Documentation. Link
These citations confirm the impact of the Okapi system and its developments on the field of information retrieval, cementing its role in advancing probabilistic retrieval models.