ISR11 An Experimental Investigation of Automatic Hierarchy Generation chapter G. Blomgren A. Goodman L. Kelly Harvard University Gerard Salton Use, reproduction, or publication, in whole or in part, is permitted for any purpose of the United States Government. T - - - Tt 1 D1 T1 v'II-3 rp - -- T 1 t Dd C. 5jk jk Tt Document-term Term-term where d d [OCRerr] [OCRerr]ILJ (C. .,c [OCRerr] MIN (C. ,c ) for = i=l ij ik S = i=1 ii ik j[OCRerr]k jk d kj d z C.. i-i i=1 ik o <S jk < 1 for j[OCRerr]k S.. is not defined and is never used JJ d d S Z ... = . C' jk [OCRerr] [OCRerr] kj i-l so that z C.. S = S kj jk [OCRerr] Cik The second step is the evaluation of relationships between pairs of terms. Choosing a `1cutoff' parameter 0 <K < 1, apply the following rules [7]: S. 1) jk <K, 5ki <K T. and T are unrelated, since the two j k terms [OCRerr]enerally are not relevant to the same documents. 2) %.k>K, Ski >K T. and T are similar, since both terms k generally are relevant to the same documents. Similar terms are called brothers.