Probabilistic Retrieval Models and Binary Independence Retrieval (BIR) Model

Synonyms

BIR model; Probabilistic model; RSJ model

Definition

Information retrieval (IR) systems aim to retrieve relevant documents while not retrieving non-relevant ones. This can be viewed as the foundation and justification of the binary independence retrieval (BIR) model, which proposes to base the ranking of documents on the division of the probability of relevance and non-relevance.

For a set r of relevant documents, and a set $\bar{r}$ of non-relevant documents, the BIR model defines the following term weight and retrieval status value (RSV) for a document-query pair “d, q”:

$${\rm birw}( t,r,\bar r)\,{\rm{: = }}\,{{P{(t|r)} \cdot P{(\bar t| \bar r)} \over {P{(t|\bar r)}} \cdot P{(\bar t|r)}}} $$

((1))

$${\rm{RSV}}_{{\rm{BIR}}} {(d,q,r,\bar r)}\,{\rm{:}}\,{\rm{ = }}\,\sum\limits_{t \in d \cap q} {{\rm \log birw}(t,r,\bar r)}$$

((2))

Here, P(t|r) is the probability that term t occurs in the relevant documents, and P(t|$\bar{r}$) is the respective probability for term tin...