Coping with Uncertainty in Schema Matching: Bayesian Networks and Agent-Based Modeling Approach

Abstract

Schema matching and mapping are an important tasks for many applications, such as data integration, data warehousing and e-commerce. Many algorithms and approaches were proposed to deal with the problem of automatic schema matching and mapping. In this work, we describe how schema matching problem can be modelled and simulated as agents where each agent learn, reason and act to find the best match in the other schema attributes group. Many differences exist between our approach and the existing practice in schema matching. First and foremost our approach is based on the paradigm Agent-based Modeling and Simulation (ABMS), while, as far as we know, all the current methods do not use ABMS paradigm. Second, the agent’s decision-making and reasoning process leverages probabilistic models (Bayesian) for matching prediction and action selection (planning). The results we obtained so far are very encouraging and reinforce our belief that many intrinsic properties of our model, such as simulations, stochasticity and emergence, contribute efficiently to the increase of the matching quality and thus the decrease of the matching uncertainty.