ISR10 Search Request Formulation chapter Joseph John Rocchio Harvard University Gerard Salton Use, reproduction, or publication, in whole or in part, is permitted for any purpose of the United States Government. rithe moQifie[OCRerr] query q1, then, is a weighte[OCRerr] vector sum of the original query vector plus the optimal vector to [OCRerr]ifferentiate the members of the set [OCRerr] from those of the set S. In other wor[OCRerr]s, q1 +5 the vector sum of q[OCRerr] plus the optimal vector for the subset of the 0 reference collection for which the user has provi[OCRerr]e[OCRerr] relevance information. If equal weight is given to the original query an[OCRerr] the optimal vector base[OCRerr] on the fee&oack information, equation (37) may[OCRerr]be written in the form: n n2 q1 = n1n2q[OCRerr] + n2 L [OCRerr] - n1 T 5[OCRerr] i=1 i=1 (3.[OCRerr]) If q1 is to be restricte[OCRerr] to a vector with only nonnegative components, the following may be used: -I f q1 f,0r.% >0 ½j 0 for q; [OCRerr]o j Fi[OCRerr]ure 3.2 provides a two-dimensional geometrical interpretation of the relevance feedback request modification process. Part (a) shows the[OCRerr]initial query q located between the 0 relevant and non-relevant document vectors. [OCRerr]he vector r-s shown in part (b) istlie optimal vector (i.e. the vector which maximizes the fun9tion C of equation (3.2) ) for differentiating the subset [OCRerr] =£r¼,r2} , from S [OCRerr] . Part (c) shows the resi[OCRerr]iitant of adding q to the normalized vector sum r-s, which results in the new 0