Scaling Probabilistic Temporal Query Evaluation
Melisachew Wudage Chekol · 2017
Open information extraction has driven automatic construction of (temporal) knowledge graphs (e.g. YAGO) that maintain probabilistic (temporal) facts and inference rules. One of the most important tasks in these knowledge graphs is query evaluation. This task is well known to be #P-hard. One of the bottlenecks of probabilistic (temporal) query evaluation is finding efficient ways of grounding the query and inference rules, to generate a factor graph that can be used for approximate query evaluation or to retrieve lineages of queries for exact evaluation. In this work, we propose the PRATiQUE (PRobAbilistic Temporal QUery Evaluation) framework for scalable temporal query evaluation. It harnesses the structure of temporal inference rules for efficient in-database grounding, i.e., it uses partitions to store structurally equivalent rules. Besides,PRATiQUE leverages a state-of-the-art Gibbs sampler to compute marginal probabilities of query answers. We report on an extensive experimental evaluation, which confirms the efficiency of our proposal.