FOA Home | UP: Mathematical Foundations

Probabilisitic retrieval

\newcommand{\si}{stochastically independent\ }

It should be clear by now that there are few logical grounds by which we could irrefutably {prove} that any document is relevant to any query. Our best hope is to retrieve relevant documents with high probability; i.e., we must about uncertaintly. Beginning in the early 1970s, the most seminal thinkers within IR have attempted to connect fundamental concepts like ``relevance'' from within IR to probability theory (see [REF327] [Cooper73] [Bookstein74] [Cooper78] [Cooper83] and other references throughout this section). As typical search engines moved past simpler Boolean retrieval to ranked retrievals [Cooper88] , and more recently as representational languages like Bayesian networks have become widespread within artificial intelligence (cf. Section §5.5.7 ), a probabilistic foundation for FOA has become an increasingly central concern. The derivation of the basic components of the Bayesian decision models presented here follows \vanR{115} quite closely; the lecture notes on probabilistic IR developed by Nobert Fuhr provide a similar treatment of this background.

Subsections

• 5.5.1 Probability Ranking Principle
• 5.5.2 Bayesian inversion
• 5.5.3 Odds calculation
• 5.5.4 Binary Independence Model
• 5.5.5 Linear discriminators
• 5.5.6 Cost analysis
• 5.5.7 Bayesian networks

Top of Page | UP: Mathematical Foundations | ,FOA Home