Meta-modelo funcional para recuperação de informação baseado em λ-cálculo

Abstract

Modeling is a topic of the central research in Information Retrieval (IR). Generic frameworks are approaches for study and development of new IR models is generic frameworks. These frameworks could be seen as formal metamodels that include a notation used to describe IR models and to make possible the research on the semantics retrieval process. Besides that, these metamodels make easier the reasoning on the characteristics and properties of information Retrieval models. In this dissertation, we offer a generic and formal functional framework, based on λ-calculus, for definition and study of the IR models, also called Functional Structure. This framework (or metamodel) allows the representation, combination formularization and equivalence comparison among IR models. In this metamodel, the IR models are represented by functions, that is, we have formally defined the components such as documents, consultations and similarity function of information Retrieval models, using the λ-calculus. This strategy has as the main element the functions concept, in this meaning it differs from the traditional models, which generally consider weights of terms, vectors, etc as beddings. When representing the IR models in a same functional language, as the λ-calculus, the identification of the arguments and values of the functions are clear. Moreover, we show examples of how to represent the classic IR models, studying the equivalence of alternative vectorial models, defined here using the Functional Structure. We also represent in this structure the following models: belief networks, sTerm and a based on ontology. By this, we verify that the representation of the IR models to the Functional Structure makes possible the construction and the combination of models, it also allows the comparison among them by its similarity (equivalence or not), through algebraic demonstrations, without carrying through experiments.