Data Bases

The problem of computer assisted literature search is often misstated as the attempt to nd, in a repository of electronic literature, a paper which, according to title and keywords, contains the desired information. This assumes that the original author was correct and complete in assigning keywords, and that the search criteria are free of preconceived expectations on the context in which the information might be found. Crossing the boundaries of disciplinary subbelds is very diicult in this manner. To avoid these problems it is necessary to categorize and classify the documents in the text data base using the full text. This will identify papers of interest, and of unexpected cross-relevance. We show the results of experiments using unsupervised classiication based on neural networks. In astronomy, several years of major journals are now accessible through the Internet. Within a very short time we will have the current body of astronomical knowledge available for computer processing. Using appropriate tools and networks we can consider the electronically available astronomical literature to be one huge text data base, consisting typically of technical/scientiic articles, written in scientiic English and using well deened terminology. In addition to improved access and timely availability electronic publications have the advantage of being searchable. Organizations like the NASA Astrophysics Data System (ADS; http://adswww.harvard.edu/) specialize in such services, freeing the user from having to read an increasingly enormous amount of material in order to nd the desired information.