Optimal Update Repair with Maximum Likelihood and Minimum Cost

Abstract

In this paper, we study the problem of update repair under integrity constraints. For a set of inconsistent data, update repair modifies the attribute values of inconsistent tuples such that the modified data no longer violate the constraints. There usually exist multiple update repairs and it is difficult to determine which one is the optimal. Most previous works prefer the one with the minimum cost (a.k.a minimum update repair) to avoid excessive modifications to the original data. However, the repair quality of minimum update repair is often far from satisfactory. We intuitively notice that the erroneous attribute values usually have low co-occurrence with the correct ones. Therefore we propose to take both the quantitative statistics and the repair cost into consideration to repair the data. Specifically, we first formalize the update repair with maximum likelihood and minimum cost problem and analyze its hardness. Then we propose an efficient approach to generate candidate repairs based on the minimum update set. Finally, a probability model together with an efficient inference approach is proposed to compute the likelihood of each candidate. Extensive experiments on real-world datasets show that our proposal can achieve higher precision and recall compared with state-of-the-art methods.