Maximally joining probabilistic data

  • Authors:

  • Benny Kimelfeld;Yehoshua Sagiv

  • Affiliations:

  • The Hebrew University of Jerusalem;The Hebrew University of Jerusalem

  • Venue:
  • Proceedings of the twenty-sixth ACM SIGMOD-SIGACT-SIGART symposium on Principles of database systems
  • Year:
  • 2007

Quantified Score

Hi-index 0.00

Visualization

Abstract

Conceptually, the common approach to manipulating probabilistic data is to evaluate relational queries and then calculate the probability of each tuple in the result. This approach ignores the possibility that the probabilities of complete answers are too low and, hence, partial answers (with sufficiently high probabilities) become important. Therefore, we consider the semantics in which answers are maximal (i.e., have the smallest degree of incompleteness), subject tothe constraint that the probability is still above a given threshold. We investigate the complexity of joining relations under the above semantics. In contrast to the deterministic case, this approach gives rise to two different enumeration problems. The first is finding all maximal sets of tuples that are join consistent, connected and have a joint probability above the threshold. The second is computing all maximal tuples that are answers of partial joins and have a probability above the threshold. Both problems are tractable under data complexity. We also consider query-and-data complexity, which rules out as efficient the following naive algorithm: compute all partial answers and then choose the maximal ones among those with probabilities above the threshold. We give efficient algorithms for several, important special cases. We also show that, in general, the first problem is NP-hard whereas the secondis #P-hard.