Incorporating window-based passage-level evidence in document retrieval

Abstract

This study investigated whether document retrieval can be improved if documents are divided into smaller sub-documents or passages and the retrieval score for these passages are incorporated in the final retrieval score for the whole document. The documents were segmented by sliding a window of a certain size across the document and extracting the words displayed each time the window stopped. A retrieval score was calculated for each of the passages extracted and the highest score obtained by a passage of that size was taken as the document’s passage-level score for that window size. A range of window sizes was tried.

The experimental results indicated that using a fixed window size of 50 words gave better results than other window sizes for the TREC-5 and TREC-6 test collections. This window size yielded a significant retrieval improvement of 24% compared to using the whole-document retrieval score (using the traditional tf^*idf weighting scheme with cosine normalisation). However, combining this window score and the whole-document retrieval score did not yield a retrieval improvement.

Using a variable window size (ranging from 50 to 400 words) yielded a retrieval improvement of about 5% over using a fixed window size of 50. Different window sizes were found to work best for different queries. If the best window size to use for each query could be predicted accurately, a maximum retrieval improvement of 42% could be obtained.

Subsequent work suggests that the usefulness of passage-level evidence in document retrieval depends on the weighting scheme and type of normalisation used in the retrieval method.