A Mutual-Information-Based Approach to Entity Reconciliation in Heterogeneous Databases

Entity reconciliation is crucial to data interoperability in heterogeneous databases. In our previous research works, we proposed an entities matching algorithm based on attribute entropy to identify the corresponding entities, which can resolve the limitations of present main approaches and improve the precision of entities matching obviously. By our further research, we find that some attributes with different importance in identifying the entities will obtain the same weights just according to attribute entropy. So in this paper we employ mutual information to quantify attribute weight due to mutual information well describes the correlation of probability distributions over two attributes. According to this idea, the final entropy computation algorithm and entity reconciliation algorithm based on mutual information are presented. The experimental results on real-world data show that our mutual-information-based approach can obtain better performance.