Um estudo sobre o uso de informações de instâncias para o casamento de esquemas no domínio de comércio eletrônico

Abstract

The Data integration task seeks to combine data from various sources and represent them in a single data set. Schema Matching is a key task to solve this problem, and is defined as the task of finding semantic correspondences between elements of two distinct schemes. Recently, this problem have been studied in the e-commerce domain, since it has great practical importance in people’s daily lives. Several methods have been proposed in the literature aiming to automate the schema matching task. These methods use different types of information, such as information on the names and structure of the elements of the analyzed schemas. In this research, we try to identify if the information of instances are more significant to the schema matching methods in the e-commerce domain. With this purpose, we verify the behavior of three schema matching methods by adding the instances information: COMA, which uses fixed heuristics to match matchers; ALMa, which uses Active Learning; And RFSM, which uses supervised machine learning. In the experiments, we noticed that by using instances information all methods presents improvements, mainly in precision and measure-f. We also verify that the methods do not require a high frequency of this information to contribute with the results.