Practical reasoning and rule mining with DatalogMTL

Thesis

DatalogMTL is a powerful temporal knowledge representation language, which has recently gained prominence in contexts involving temporal data. Its applications extend to areas like ontology-based query answering, network flow anomaly detection, equipment malfunction monitoring, and streaming reasoning, showcasing its versatility and effectiveness in handling time-sensitive information. Reasoning in DatalogMTL is, however, of high complexity, namely ExpSpace-complete and PSpace-complete with respect to data size, which makes reasoning in data-intensive applications challenging. Thus, theoretical research has focused on establishing a suitable trade-off between expressive power and complexity of reasoning, by identifying lower complexity fragments of DatalogMTL and studying alternative semantics with favourable computa- tional behaviour. The design and implementation of practical reasoning algorithms for DatalogMTL remains, however, a largely unexplored area—something that has so far prevented its widespread adoption in applications. In this thesis, our primary focus lies in the exploration of practical temporal reasoning algorithms in DatalogMTL. Our proposed algorithms were implemented within a system named MeTeoR. The results of our evaluation underscore the practical viability and effectiveness of our approach. Besides, we have developed a universal framework for learning rules written in DatalogMTL which capitalises on the availability of increasingly mature non-temporal rule learners. We have implemented the approach in a system MTLearn which is compatible with any Datalog rule learner, as well as with a wide range of strategies for scoring the output temporal rules. Our evaluation demonstrates its competitive performance, on par with state-of-the-art deep learning models, while also providing easily interpretable results.

Authors: Wang, D

• DOI: 10.5287/ora-nzdyprbyn
• Type of award: DPhil
• Level of award: Doctoral
• Awarding institution: University of Oxford
• Language: English
• Keywords: rule mining; temporal reasoning; DatalogMTL
• Subjects: Computer Science