Skip to Main Content
Identifying database corresponding attributes in schema matching plays a key role in data integration in heterogeneous databases. Most of current approaches mainly use schema information of attribute. Little research has attempted to fully explore the use of data content. This paper introduces a novel schema matching algorithm based on data content, which has two-step process. First, through the analysis of the data pattern, we train a set of neural networks which used for calculating candidate matching pairs. Then we apply a rule-based algorithm to filter the candidate pairs and get correct matching result. The experiment result based on real data shows our proposed approach can improve the precision and recall of schema matching obviously. The approach can either be used independently or work together with other schema matching methods.