ChronoQG: New Benchmark for Temporal Knowledge Graph Question Generation

This research addresses a significant gap in the field of Knowledge Graph Question Generation (KGQG) by introducing ChronoQG, the first benchmark specifically designed for Temporal Knowledge Graph Question Generation (TKGQG). Current KGQG benchmarks primarily rely on static knowledge graphs, failing to account for the temporal dimensions of facts. This limitation means they cannot adequately assess whether generated questions accurately reflect temporal validity, event ordering, or answer-determining temporal constraints. ChronoQG proposes a comprehensive framework for TKGQG, where generated questions must be faithful to both the supporting subgraph and the temporal constraints necessary to identify the correct answer. The framework incorporates a detailed temporal-constraint taxonomy, topology-temporal subgraph sampling, and trace-grounded question generation to ensure the construction of temporally faithful questions. The authors used this framework to create four benchmark datasets from diverse temporal knowledge graphs, resulting in 16,011 verified questions. Evaluations of representative LLM-based KGQG methods and prompting baselines across various TKGQG settings (including temporal-constraint counts, topological templates, and constraint types) revealed a clear deficiency: existing methods struggle significantly to preserve temporal constraints, particularly in multi-constraint scenarios and with more complex temporal constraint types. This establishes ChronoQG as a challenging new testbed for advancing temporally faithful question generation.