Mining : Origins of Database Tomography and Multi - Word Phrase Clustering

This report initially describes the motivations for co-word analysis in support of research policy formulation and research implementation evaluation. It places coword analysis in perspective to other co-occurrence techniques such as cocitation and co-nomination analyses. It then traces the origins of co-word analysis in computational linguistics, describes in detail the development of co-word analysis for research evaluation, and concludes by presenting a new approach to co-word analysis for research evaluation (Database Tomography). The report shows that this new approach to co-word analysis, which requires no index or key words but deals with text directly, is a useful tool for scanning large bodies of text. It can identify pervasive thrust areas and their interrelationships, and serve as a starting point for further in-depth analysis of the text. Its value increases as: 1) the size of text increases and 2) the breadth of topical areas covered by the text increases beyond the expertise of a moderate number of panels of experts. A single link clustering example is shown that represents the first use of multi-word technical phrases in modern clustering.