TEILP: Time Prediction over Knowledge Graphs via Logical Reasoning

TL;DR

TEILP introduces a temporal logical reasoning framework for time prediction on knowledge graphs by converting TKGs into a Temporal Event Knowledge Graph (TEKG) and learning temporal logical rules with associated conditional probability densities. A differentiable random-walk mechanism guides rule-grounding over TEKG, producing time predictions via Gaussian/exponential mixture models tied to rule patterns and query intervals. The approach yields empirically superior performance over state-of-the-art baselines across five benchmarks and demonstrates robustness in low-data, imbalanced, and future-event forecasting scenarios, while providing human-readable explanations through learned rules. Overall, TEILP bridges inductive logical reasoning and probabilistic time modeling to deliver accurate, interpretable time predictions in dynamic relational data.

Abstract

Conventional embedding-based models approach event time prediction in temporal knowledge graphs (TKGs) as a ranking problem. However, they often fall short in capturing essential temporal relationships such as order and distance. In this paper, we propose TEILP, a logical reasoning framework that naturally integrates such temporal elements into knowledge graph predictions. We first convert TKGs into a temporal event knowledge graph (TEKG) which has a more explicit representation of time in term of nodes of the graph. The TEKG equips us to develop a differentiable random walk approach to time prediction. Finally, we introduce conditional probability density functions, associated with the logical rules involving the query interval, using which we arrive at the time prediction. We compare TEILP with state-of-the-art methods on five benchmark datasets. We show that our model achieves a significant improvement over baselines while providing interpretable explanations. In particular, we consider several scenarios where training samples are limited, event types are imbalanced, and forecasting the time of future events based on only past events is desired. In all these cases, TEILP outperforms state-of-the-art methods in terms of robustness.

Figures (8)

• Figure 1: An example TKG (left) and the corresponding TEKG (right). The first row are the original versions, and the second row are the enhanced versions.
• Figure 2: The illustration of rule-based time prediction.
• Figure 3: The conditional probability distribution for the query interval given by TEILP.
• Figure 4: The conditional probability distribution given by TEILP.
• Figure 5: The conditional probability distribution for the query interval given by TEILP.