Information Retrieval (IR) models come in two main forms:
vector-based and probabilistic. Each form is an expressive framework in
which many models can be expressed. What is the common ground, and how are
models related? I will report on results how to express major IR concepts
(term frequencies, IR models, authorities, evaluation) in a general matrix
framework (Roelleke/etal:IPM:2006), and how to relate probabilistic models
(binary independent retrieval, language modelling, Poisson) in a parallel
derivation where event spaces and background models are explicit
(Roelleke/Wang:SIGIR:2006).
We investigated the underlying general concepts since we wish to support
the high-level engineering of customised IR models in our DB+IR framework
HySpirit, a probabilistic deductive DB+IR system (Fuhr/Roelleke:TOIS:1997,
Fuhr/etal:SIGIR:1998, Roelleke/etal:TREC:2005, www.apriorie.co.uk), and I
will demonstrate some HySpirit features.