IBSimilarity

Provides a framework for the family of information-based models, as described in Stéphane Clinchant and Eric Gaussier. 2010. Information-based models for ad hoc IR. In Proceeding of the 33rd international ACM SIGIR conference on Research and development in information retrieval (SIGIR '10). ACM, New York, NY, USA, 234-241.

The retrieval function is of the form RSV(q, d) = Σ -xqw log Prob(Xw ≥ tdw | λw), where

• xqw is the query boost;

• Xw is a random variable that counts the occurrences of word w;

• tdw is the normalized term frequency;

• λw is a parameter.

The framework described in the paper has many similarities to the DFR framework (see DFRSimilarity). It is possible that the two Similarities will be merged at one point.

To construct an IBSimilarity, you must specify the implementations for all three components of the Information-Based model.

1. Distribution: Probabilistic distribution used to model term occurrence

• Log-logistic: Smoothed power-law

2. Lambda: λw parameter of the probability distribution

• LambdaDF: Nw/N or average number of documents where w occurs

• LambdaTTF: Fw/N or average number of occurrences of w in the collection

3. Normalization: Term frequency normalization

• Any supported DFR normalization (listed in DFRSimilarity)

See also