Optimal Linear Ordering of Information Items

Abstract

Efficient search of items in a store of information, such as books in a library, abstracts of articles, information in a computer data bank or the like, requires that the items most closely connected in content be close together, in order that a person looking for some particular information may concentrate his search on as small a part of the store as possible. The degree of “connectedness” between two items i and j can be expressed in terms of a correlation index nij. This paper discusses methods of estimating these indices. Once the correlation indices are known for each pair of items in the store, it is possible to assign a position xi for each item i, along a linear classification scale, such that the larger the nij between items i and j the closer the two items are on the scale. Such a scale indicates the optimal ordering of books on library shelves or of items in any other data collection. It also provides means for assigning optimal subject descriptors for a computer-based information store. A procedure is described whereby the values of the xi's may be determined in terms of the nij's. When the items in the store tend to “clump together” in classes, with much smaller correlation between items of different classes than those between items of the same class, an approximation technique may be used to locate the relative positions of the classes along the scale, as well as the distribution of the x's within each particular class. An example is given of these techniques.