Taming Data Explosion in Probabilistic Information Integration

Data integration has been a challenging problem for decades. In autonomous data integration, i.e., without a user to solve semantic uncertainty and conflicts between data sources, it even becomes a serious bottleneck. A probabilistic approach seems promising as it does not require extensive semantic annotations nor user interaction at integration time. It simply teaches the application how to generically cope with uncertainty. Unfortunately, without any world knowledge, uncertainty abounds as almost everything becomes (theoretically) possible and maintaining all possibilities produces huge volumes of data. In this paper, we claim that simple and generic knowledge rules are sufficient to drastically reduce uncertainty, hence tame data explosion to a manageable size.