An experiment in automatic indexing

This report describes a method of indexing documents which is based on the assumptions, (1) that a subset of the words in a document can be an effective index to that document and, (2) that this subject can be approximated by selecting those words from the document whose frequencies are statistically unexpectedly high. The results of the experiment are encouraging, although not definitive because any index set chosen must be tested by using it for retrieval from a large collection.