GB1343867A - Data processing system - Google Patents

Abstract

1343867 Comparing data INTERNATIONAL BUSINESS MACHINES CORP 21 April 1972 [17 June 1971] 18555/72 Heading G4A The degree of similarity between first and second character strings is determined by forming a matrix M, the elements αij of which have the value "1" when the Ith character of the first string is identical to the Jth character of the second string and the value "0" otherwise. In each row, or column of the matrix (or both) only the "1" value nearest the diagonal of the matrix is retained, the X and γ co-ordinate positions of these "1" values in the matrix being used to compute the standard correlation co-efficient to measure the similarity between the character strings. For example, the terms ANNE and ANNIE are compared by deriving a five by four matrix which, after rejecting all the "1's" in each row except that nearest the diagonal, becomes i.e. there are "1" values in the X, Y co-ordinate positions (1, 1), (2, 2), (3, 3) and (5, 4). The correlation co-efficient given by where N equals the number of elements in the matrix (that is 20) is then computed to obtain a correlation co-efficient of 0À9827. This comparison method is used in a processing system in which an operator makes a request using input/ output device 11 (Fig. 3). The request is compared with all the questions stored in a table Q in main store 13 to derive a correlation coefficient as described above for each question, the co-efficients being then stored in a list "SCORE". Associated with each question is a threshold level for the correlation co-efficients stored in list "TH", these being compared with the computed co-efficients and a zero entered in the list "SCORE" when the threshold level exceeds the co-efficient. The highest co-efficient is then read into a register "MAX", the number of the corresponding question being entered into a register "IMAX". The answer in a table "ANS" associated with the selected question is then printed out. If, even with a high correlation co-efficient, the answer is wrong, the user types a query which results in the processor asking the user if the answer is correct, this check being effected automatically if the correlation co-efficient is low. If the user indicates that the answer is wrong the threshold for the selected question is increased to above the computed correlation co-efficient and the processor requests the correct answer which is entered into the answer table, the question being stored in the question table. If the user requests the processor to try again the question with the next highest co-efficient is selected if it exceeds its threshold level. The system may be used in adaptive query systems such as passenger reservation systems or language processors.