The articles by Egghe, by Rousseau and Van Hooydonk, and by Kokol and Kokol are bibliometric in nature, while Persin et al. deals with efficient data structures for information retrieval. Wolfram examines a hypertext structure for information retrieval.
Dillon and Schaap are concerned with readers' ability to recognize the structural portion of a paper being read. Forty-eight subjects viewed paragraphs of text from published papers and allocated them to one of four classes: introduction, method, results, or discussion. Experienced readers are able to locate themselves more quickly and correctly.
Persin, Zobel, and Sacks-Davis provide some important insight into data structure for retrieval. By sorting the lists of document numbers and word counts associated with any term in an index file by the count rather than by the document number, the ability to compress the list by using run lengths from the previous document number is lost. However the need for an accumulator of similarity values for each document with other than a zero similarity value is avoided. Using thresholds based on the similarity ofthe currently most similar document, and computed before the processing of the list for each term, document accumulators are created, ignored, or augmented, or not, if already in existence. Small partial similarities are unlikely to change the final ranking and documents yielding such values are ignored at considerable memory saving. The sort by occurrence count means considerable reduction in processing time since the tails of the long count in document lists need not be processed. If the maximum in document frequency in the list is stored with the term, it is possible to avoid reading the list for some terms. One can regain some compression by sorting documents in the list with the same frequency by document number. Tests show no degradation in retrieval effectiveness and would permit ranked retrieval on considerably smaller machines.
Wolfram's HyperLynx system is tested on nearly 3000 NTIS document records on library and information science from 1989 to '9 1. Initial entry is through author, title, and descriptor indices. Each search term