Computer bibliometrics for journal classification

摘要

Data on article distributions over journal titles and subject subdivisions of a selected field could be extracted (on- or offline) from every bibliographic file including a classification scheme. On the basis of such data, journals could be subdivided into specialized, average, or general using an appropriate measure of dispersion (or concentration). To this end, Pratt's absolute measure of dispersion q = σr∗ Fr, is suggested, where the Fr, represent the relative frequencies (in descending order) of articles from a given journal in the unidentified subject subdivisions with rank r = 1, 2, …, n. In order to separate specialized and general journals from average ones, it is assumed that each empirical q-value has a ‘random’ deviation d. A w-statistic is introduced to test whether the q-values differ significantly from the average qav on a specified confidence level, where w = (q — qav)/d. In addition, a subject relative measure of dispersion Q is used to determine which subdivisions are favored by which journals. Another w-statistic is proposed to test whether the Q-values differ significantly from Qs (absolute measure of dispersion of the subject distribution). This bibliometric technique is applied to data from the 1984 INSPEC file. The results could help library staff and information scientists in classifying journals according to the two measures of article dispersion over subject subdivisions.