Time-aware Reasoning in Uncertain Knowledge Bases

Abstract

Time information is ubiquitous on the Web, and consideringtemporal constraints among facts extracted from the Web is key forhigh-precision query answering over time-variant factual data. In thispaper, we present a simple and efficient representation model for timedependentuncertainty in combination with first-order inference rulesand recursive queries over RDF-like knowledge bases. In the spirit ofdata lineage, the intensional (i.e., rule-based) structure of query answersis reflected by Boolean formulas that capture the logical dependencies ofeach derived answer fact back to its extensional roots (i.e., base facts).Our approach incorporates simple weight aggregations for begin, end andduring evidences for base facts, but also generalizes the common possibleworldssemantics known from probabilistic databases to histogram-likeconfidence distributions for derived facts. In particular, we show thatadding time to the latter probabilistic setting adds only a light overheadin comparison to a time-unaware probabilistic setting.