Some issues in efficient implementation of a vector based model for document retrieval

Issues affecting efficient computational kernels are addressed. Recent work by Howell, et al discusses how a term document matrix can be stably approximated by a low rank matrix. This paper surveys some of the problems in getting a portable efficient query implementation. It reflects the efforts of a graduate seminar class in assembling a survey of the available literature, as well as developing some of the required computational kernel. One target platform is a cluster of SGI O2 200 MHz workstations (connected by a 100 MBit switch) using message passing under the MPICH implementation of MPI. Simultaneously, we have been working on Pentium processors with a view to implementing document retrieval algorithms on a Beowulf computer.