Automatic extraction of relationships between terms by means of Kohonen’s algorithm Vicente P. Guerrero a, *, Fe ´lix Moya-Anego ´n b , Victor Herrero-Solana b a Facultad de Biblioteconomı ´a y Documentacio ´n (Alcazaba de Badajoz), Universidad de Extremadura, 06071 Badajoz, Spain E-mail address: vicente@alcazaba.unex.es (V.P. Guerrero) b Facultad de Biblioteconomı ´a y Documentacio ´n, Universidad de Granada, Colegio Ma ´ximo de la Cartuja, 18071 Granada, Spain Abstract This article describes a method of finding the contextual relationships among different terms in a database. First, the vector model is used to represent the terms as vectors according to which documents they appear in. Second, these vectors are used as the input to a Kohonen network, which organizes them topologically. This organization, in turn, generates term clusters arranged on a grid, so that each term is not only related to the others in its own cluster but also to those of neighboring clusters. D 2002 Elsevier Science Inc. All rights reserved. The exponential growth of information has been a topic of concern and interest for more than 25 years (Price, 1973). The availability of electronic information and the process of digitalization have contributed in large part to this growth, with the transformation of documents ‘‘based on atoms to ones based on bits’’ (Negroponte, 1995). Computer use is, of course, not restricted to editorial production; it is present in all aspects of life — from the workplace, where there is often one computer per person, each of whom is generating new documents, to the home, where, increasingly, people not only have a computer but multimedia equipment as well. There is also the distribution of information via the so-called information highway and the effect of the increasingly lower cost of storage media. One is thus in the midst of a developing environment of electronic information that can be accessed automatically. Another consideration is the diversification of media, which has the collateral effect of producing a greater amount of nonnormalized information (e.g., images, sound, and text). 0740-8188/02/$ – see front matter D 2002 Elsevier Science Inc. All rights reserved. PII:S0740-8188(02)00124-X * Corresponding author. Library & Information Science Research 24 (2002) 235 – 250