A statistical learning approach to automatic indexing of controlled index terms

A statistical learning approach to assigning controlled index terms is presented. In this approach, there are two processes: (1) The learning process and (2) the indexing process. The learning process constructs a relationship between an index term and the words relevant and irrelevant to it, based on the positive training set and negative training set, which are sample documents indexed by the index term, and those not indexed by it, respectively. The indexing process determines whether an index term is assigned to a certain document, based on the relationship constructed by the learning process, and the text found in the document. Furthermore, a learning feedback technique is introduced. This technique used in the learning process modifies the relationship between an index term and its relevant and irrelevant words to improve the learning performance and, thus, the indexing performance. Experimental results have shown that the statistical learning approach and the learning feedback technique are practical means to automatic indexing of controlled index terms.