The development and evaluation of a full-text drugs database Martindale Online
Reynolds, J. E. F. (1987). The development and evaluation of a full-text drugs database Martindale Online. (Unpublished Doctoral thesis, The City University)
Abstract
Martindale Online is a full-text database on drugs produced from a structured neutral database that is also used to produce a print product. Special characteristics of the database include a hierarchical record structure and a facility for linking records within the same hierarchy. The development of this database is described.
Investigation at the development stage indicated a need to index the database and this was carried out using descriptors from a specially designed thesaurus. Ta evaluate the effect of this indexing, three information pharmacists selected 98 queries for an assessment of retrieval effectiveness; they and the author formulated sets of search statements that were used to search the file in several different ways. It was found that searching the indexed database via descriptors and free text (when appropriate) produced significantly better results, as judged by scores that incorporated precision and recall, than searching either the indexed or the unindexed database solely in a free-text manner.
As there was evidence that searchers were slow to make use of the descriptors, highly structured search statements were created for each query using all the details from the relevant sections of the thesaurus and these statements were tested on the unindexed database. While this test produced some conflicting results, it did suggest that as far as major relevance was concerned such a method of searching might be effective with Martindale Online and is worth exploring further, especially with a view to producing a front-end system.
Detailed failure analysis was carried out on the searches performed in the recommended manner. With the information pharmacists’ search statements the database was operating at a recall ratio of 60.2 for all relevant records (69.3 for records of major relevance); with the author’s statements the recall ratio was 65.4 (73.2 for major relevance). Corresponding precision ratios were 63.5 (58.3 for major relevance) for the information pharmacists and 67.5 (59.6) for the author.
The largest cause of both recall and precision failure was in limitations of the search statements whether produced by the information pharmacists who had varied experience of Martindale Online, or by the author who has a detailed knowledge of the system and the contents. Limitations in the indexing also accounted for both types of failure; account has already been taken of these limitations and modifications have been made to some of the indexing guidelines.
Download (118MB) | Preview
Export
Downloads
Downloads per month over past year