A COMPARISON OF KEYWORD-IN-CONTEXT (KWIC) INDEXING TO MANUAL INDEXING.

The Knowledge Availability Systems Center at the Univ. of Pittsburg evaluated, by three different methods, the effectiveness of an information retrieval system indexed by 2 different methods. The sample for the evaluation was from questions submitted to a Spin-off project of the National Aeronautics and Space Agency on providing a facility to disseminate NASA-sponsored research into the local area economy. The first evaluation was based on percentages of correct retrievals and correct rejections. The KWIC and manually indexed systems were 56.39 and 68.91 effective, respectively. The second evaluation was based on conditioned probabilities KWIC was even less effective when documents retrieved and not retrieved were looked at, however KWIC exhibited a 0.938 probability that documents not pertinent to the question would be retrieved. By allowance of a little more noise, KWIC could become a more effective indexing technique. The third evaluation on a document-bydocument analysis basis provided results similar to those of the second method. Although KWIC is not regarded as very effective for retrieving pertinent information. There is little doubt that it can be effective in very specialized situations.