Efficiency/Effectiveness Trade-offs in Query Processing (from Theory into Practice Workshop, 1998 SIGIR Conf.) David Hawking CSIRO Mathematical and Information Sciences G P O B o x 664, C a n b e r r a A C T 2601, A u s t r a l i a David. Hawking@cmis.csiro. au * O c t o b e r 31, 1998 Introduction A particular variant of Okapi [Robertson et al. 1994] relevance scoring implemented in PADRE [HaWking et al. 1997] achieved best overall average precision in the Long-Topic Adhoc category of the 1997 T R E C conference. [Voorhees and Harman 1997] Here, this particular implementation is used as the starting point for a study of the efficiency/effectiveness trade-offs inherent in a number of optimisation techniques, listed in Section 4. Overview of P A D R E Query Processing (Using Okapi Model) In the present experiments, queries were processed as follows: Each query term was stemmed and then looked up (using binary search) in a sorted term dictionary which gives the offset and length of the term's posting list within the inverted file as well as its dr. A posting is a (document_number, value) pair, where value may
/lp/association-for-computing-machinery/efficiency-effectiveness-trade-offs-in-query-processing-from-theory-A5MN0mnQ8v