© 2000 by British Computer Society
Reversing the Error-Correction Scheme for a Fault-Tolerant Indexing
1 Department of Computer Science, The George Washington University, Washington D.C. 20052, USA Email: berkov@seas.gwu.edu
The paper presents an innovative approach to approximate matching of multi-attribute objects based on reversing of the conventional scheme of error-correction coding. Approximate matching problems primarily arise in information retrieval systems storing fuzzily described items and operating with nebulous searching criteria. Applying a decoding procedure to binary vectors corresponding to multi-attribute objects, the obtained message words can be used as indices of these objects, so certain mismatches in their attributes can be tolerated. This technique is mostly efficient when the error-correcting code is perfect. Thus, the simplest practical choice for the realization of this technique is based on the so-called perfect Golay code that maps 23-bit vectors into 12-bit message words. In this case, two different 23-bit vectors at a Hamming distance of 2 would have some common 12-bit indices. This allows the organizing of a direct retrieval of a neighbourhood of 23-bit vectors with up to two mismatches from a given key. The proposed technique employs a reasonable redundancy and can trade utilization of extra memory for the speed and range of searching.
Received 23 March, 1999. Revised 15 November, 1999.