ABSTRACT
This paper describes a simple way of adapting the BM25 ranking formula to deal with structured documents. In the past it has been common to compute scores for the individual fields (e.g. title and body) independently and then combine these scores (typically linearly) to arrive at a final score for the document. We highlight how this approach can lead to poor performance by breaking the carefully constructed non-linear saturation of term frequency in the BM25 function. We propose a much more intuitive alternative which weights term frequencies <i>before</i> the non-linear term frequency saturation function is applied. In this scheme, a structured document with a title weight of two is mapped to an unstructured document with the title content repeated twice. This more verbose unstructured document is then ranked in the usual way. We demonstrate the advantages of this method with experiments on Reuters Vol1 and the TREC dotGov collection.
- David Carmel, Yoelle S. Maarek, Matan Mandelbrod, Yosi Mass, and Aya Soffer. Searching xml documents via xml fragments. In Proceedings of the 26th annual international ACM SIGIR conference on Research and development in informaion retrieval, pages 151--158. ACM Press, 2003. Google ScholarDigital Library
- Nick Craswell and David Hawking. Overview of the TREC-2002 Web track. In TREC 2002, 2003.Google Scholar
- INEX. Initiative for the evaluation of XML retrieval (INEX), http://inex.is.informatik.uni-duisburg.de:2003.Google Scholar
- Evangelos Kotsakis. Structured information retrieval in XML documents. In SAC 2002, volume 1-58113-445-2/02/03, Madrid, Spain, 2002. ACM. Google ScholarDigital Library
- Mounia Lalmas. Uniform representation of content and structure for structured document retrieval. Technical report, Queen Mary and Westfield College, University of London, 2000.Google Scholar
- S H Myaeng, D-H Jang, M-S Kim, and Z-C Zhoo. A flexible model for retrieval of SGML documents. In W B Croft, A Moffat, C J van Rijsbergen, R Wilkinson, and J Zobel, editors, SIGIR'98: Proceedings of the 21st Annual International ACM SIGIR Conference on Research and Development in Information Retrieval, pages 138--145. ACM Press, 1998. Google ScholarDigital Library
- N.Craswell, D.Hawking, A.McLean, T.Upstill, R.Wilkinson and M.Wu. TREC 12 web track at CSIRO. In TREC 2003, 2004.Google Scholar
- Hongbo Xu, Zhifeng Yang, Bin Wang, Bin Liu, Jun Cheng, Yue Liu, Zhe Yang, Xueqi Cheng and Shuo Bai TREC-11 experiments at CAS-ICT: Filtering and Web. In TREC 2002, 2003.Google Scholar
- Lide Wu, Xuanjing Huang, Junyu Niu, Yingju Xia, Zhe Feng and Yaqian Zhou. FDU at TREC2002: Filtering, Q&A, Web and Video tasks. In TREC 2002, 2003.Google Scholar
- Einat Amitay, David Carmel, Adam Darlow, Ronny Lempel and Aya Soffer. Topic distillation with knowledge agents. In TREC 2002, 2003.Google Scholar
- Abdur Chowdhury, Mohammed Aljlayl, Eric Jensen, Steve Beitzel, David Grossman and Ophir Frieder. Linear combinations based on document structure and varied stemming for Arabic retrieval. In TREC 2002, 2003.Google Scholar
- Nie Yu, Ji Donghong and Yang Lingpeng. LIT at TREC-2002: Web track. In TREC 2002, 2003.Google Scholar
- Shuang Liu, Clement Yu and Wensheng Wu. UIC at TREC-2002: Web track. In TREC 2002, 2003.Google Scholar
- Jacques Savoy and Yves Rasolofo. Report on TREC-11 experiment: Arabic, Named Page and Topic Distillation searches. In TREC 2002, 2003.Google Scholar
- Paul Ogilvie and Jamie Callan. Combining document representations for known item search. In Proceedings of the 26th Annual International ACM SIGIR Conference on Research and Development in Information Retrieval (SIGIR 2003), 2003. Google ScholarDigital Library
- Benjamin Piwowarski and Patrick Gallinari. A machine learning model for information retrieval with structured documents. In Petra Perner, editor, Machine Learning and Data Mining in Pattern Recognition (MLDM'03), pages 425--438, Leipzig, Germany, July 2003. Springer Verlag. Google ScholarDigital Library
- ReutersI. Reuters corpus volume 1, http://about.reuters.com/researchandstandards/corpus/index.asp.Google Scholar
- S E Robertson and S Walker. Some simple effective approximations to the 2-Poisson model for probabilistic weighted retrieval. In W B Croft and C J van Rijsbergen, editors, SIGIR '94: Proceedings of the 17th Annual International ACM SIGIR Conference on Research and Development in Information Retrieval, pages 345--354. Springer-Verlag, 1994. Google ScholarDigital Library
- Ross Wilkinson. Effective retrieval of structured documents. In Research and Development in Information Retrieval, pages 311--317, 1994. Google ScholarDigital Library
Index Terms
- Simple BM25 extension to multiple weighted fields
Recommendations
BERT-based Dense Retrievers Require Interpolation with BM25 for Effective Passage Retrieval
ICTIR '21: Proceedings of the 2021 ACM SIGIR International Conference on Theory of Information RetrievalThe integration of pre-trained deep language models, such as BERT, into retrieval and ranking pipelines has shown to provide large effectiveness gains over traditional bag-of-words models in the passage retrieval task. However, the best setup for ...
When documents are very long, BM25 fails!
SIGIR '11: Proceedings of the 34th international ACM SIGIR conference on Research and development in Information RetrievalWe reveal that the Okapi BM25 retrieval function tends to overly penalize very long documents. To address this problem, we present a simple yet effective extension of BM25, namely BM25L, which "shifts" the term frequency normalization formula to boost ...
Weighted shingling: an adaptation of shingling for weighted shingles
IIT'09: Proceedings of the 6th international conference on Innovations in information technologyBroder's shingling is one of the state-of-the-art approaches in detecting near-duplicate documents. Prior evaluations of this method have shown that document-pairs which have different main content but have a large amount ofsimilar unimportant details ...
Comments