Toward the Principled Utilization of Contextual Evidence within the Information Retrieval Process: Investigating the theoretical foundation of Language Model Based Information Retrieval

The language modelling approach to Information Retrieval (IR) has generated much interest in the field since its conception in 1998[73]. However, some serious questions have been asked about the integrity of the language modelling approach. Specifically, it does not model relevance explicitly, unlike traditional probabilistic models of IR such as the Binary Independence Model[93]. Instead, it relies upon several underlying assumptions which are touted as being correlated with relevance. In this document, we provide a review of current state of the art language modelling approaches to IR and discuss the conjecture surrounding the language modelling approach. We then provide a study which analyzes the relationship between perplexity and Average Precision that underpins the language modelling approach. We conclude this document by detailing some potential future directions of the Ph.D, the expected contributions of the work and the proposed timetable.