Text Passage Retrieval Based on Colon Classification: Analysis of Results

A set of experiments was conducted to determine the suitability of the Colon Classification as a foundation for the automated analysis, representation, and retrieval of primary information from the full text of documents.

Full text data bases were created in two subject areas and queries solicited from specialists in each area. An automated and queries solicited from specialists in each area. An automated full text indexing system, along with four automated passage retrieval systems, was created to test the various features of the Colon Classification. Two Boolean-based systems and one simple word occurrence system were created in order to Compare the retrieval results against types of systems which are in more common use.

The results of these experiments are discussed in the context of a detailed analysis of retrieval failures of four of the retrieval systems the simple word occurrence system, one Boolean-based system, and two of the Colon Classification-based systems. The failure analysis attempted to determine why a system. retrieved irrelevant paragraphs and did not retrieve relevant paragraphs.

Although it was found that the Colon Classification-based systems did not perform  significantly better than the other systems, the analysis of retrieval failures has identified a set of procedures which should improve this performance. In addition, this analysis has identified certain interesting areas for research and development.