TempoQR: Temporal Question Reasoning over Knowledge Graphs

Knowledge Graph Question Answering (KGQA) involves retrieving facts from a (KG) using natural language queries. A KG is curated set of consisting entities linked by relations. Certain include also temporal information forming Temporal (TKG). Although many questions involve explicit or implicit time constraints, question answering (QA) over TKGs has been relatively unexplored area. Existing solutions are mainly designed for simple that can be answered directly single TKG fact. This paper puts forth comprehensive embedding-based framework complex TKGs. Our method termed reasoning (TempoQR) exploits embeddings to ground the specific and scope it refers to. It does so augmenting with context, entity time-aware employing three specialized modules. The first computes textual representation given question, second combines involved in third generates question-specific embeddings. Finally, transformer-based encoder learns fuse generated representation, which used answer predictions. Extensive experiments show TempoQR improves accuracy 25--45 percentage points on state-of-the-art approaches generalizes better unseen types.