EVALUATION OF THE SCREENING STAGES OF THE SHEFFIELD RESEARCH-PROJECT ON COMPUTER-STORAGE AND RETRIEVAL OF GENERIC CHEMICAL STRUCTURES IN PATENTS

An evaluation is given of the search strategies used in the screening stages of the Sheffield generic structure system. The results of sever al searches are presented; a detailed account is given of their perfor mance together with a discussion of outstanding problems and weaknesse s. The screening stages are of two types, a fast bitscreening stage us ing inverted files of bitstrings and a slower but more discriminating stage which operates on files of reduced graphs. Bitscreens are used t o represent the presence, absence, or possible presence of features at tributable to the structure. These features are of two types: fragment descriptors and ring descriptors. The results of searches using these screening stages are encouraging. The diversity of the methodology us ed is reflected in the results; the strengths of each method are shown to resolve the weaknesses of others. The final stage of search, the r efined search, is introduced. This is a localized atom-level search wh ich investigates, in greater detail, the mappings between reduced grap h nodes which result from the reduced graph screening stage. This is t he most discriminating search stage and is therefore expected to be hi ghly efficient.