Relevance odds of retrieval overlaps from seven search fields

It is shown that duplicate documents retrieved by the use of terms from any two of the fields would have higher odds of being judged relevant than those retrieved by only one of the Fields.

Abstract

Data contained in a 1982 paper were analyzed in terms of relevance odds of common items retrieved by searching any two content-bearing search fields. While the 1982 study compared the relative retrieval performance of 7 search fields, the present study shows that duplicate documents retrieved by the use of terms from any two of the fields would have higher odds of being judged relevant than those retrieved by only one of the fields. Sixty-three relevance odds were computed using the log cross product technique. The highest relevance odds were associated with common items retrieved from assigned descriptors and from truncated free-text terms from either the title or abstract fields; their relevance odds were 19 to 2 in favor of overlaps. Overlap retrieval could be considered a strategy for high precision searching.