Schema Mapping Using Hybrid Ripple-Down Rules

Schema mapping is essential to manage schema heterogeneity among different sources. Schema mapping can be conducted by using machine learning algorithms or by knowledge engineering approaches. These two approaches have advantages and disadvantages. The machine learning approaches can learn their model using the data, but they are static, so they cannot be modified to reflect the domain data changes. Inversely, the knowledge engineering approaches need domain experts, but they can be modified by reflecting the domain data changes. In order to exploit the advantages of both approaches and reduce the limitations, we propose a hybrid approach, called Hybrid-RDR, which combines a machine learning algorithm with ripple-down rules (RDR), an incremental knowledge engineering approach. A model is constructed by a decision tree algorithm and then it is extended by adding rules incrementally. This approach achieves higher performance in terms of precision, recall and F-measure compared to the machine learning algorithm. This significantly reduces the effort for classifying the related schemas one by one by manually creating rules and it is possible to modify the knowledge base by adding rules without creating model again if decision tree gives wrong classifications whenever the schema data changes over time.