Information Retrieval using Poisson Query Generation Model

This paper proposes a query generation model using Poisson distribution. Most existing models use multinomial distribution and score documents based on query likelihood that was computed by a query generation probabilistic model. It can be seen that the new model and the existing multinomial models are equivalent but behave differently in smoothing methods. It is found that Poisson model has several advantages over the multinomial model like naturally accommodating �per-term smoothing� and allowing more accurate background modeling. The paper presents several type of the above described model corresponding to different methods, and evaluates them.