TL;DR
UniKGQA collapses KG subgraph retrieval and answer reasoning into a single UniModel with shared PLM parameters, outperforming pipeline baselines on WebQSP, CWQ, and MetaQA through joint end-to-end training.
Summary
Multi-hop knowledge graph question answering (KGQA) traditionally separates subgraph retrieval and answer reasoning into a pipeline, causing error propagation between stages. UniKGQA collapses both into a single UniModel built on a pretrained language model that shares parameters across retrieval (finding relevant KG paths) and reasoning (reading paths to produce answers). Joint end-to-end training allows retrieval signals to inform reasoning and vice versa. On WebQSP, CWQ, and MetaQA benchmarks, UniKGQA outperforms pipeline and retrieval-only baselines, confirming that joint optimization of retrieval and reasoning is superior to staged approaches.
Key contributions
- Proposes UniModel, a single PLM-based model that shares parameters across subgraph retrieval and answer reasoning for KGQA.
- Eliminates error propagation between retrieval and reasoning stages through joint end-to-end training.
- Frames subgraph retrieval as path scoring, enabling differentiable integration with the reasoning component.
- Outperforms pipeline baselines on WebQSP, CWQ, and MetaQA multi-hop benchmarks with a smaller overall model footprint.
When to cite
- When arguing that unified retrieve-and-reason models outperform modular pipelines for multi-hop KGQA.
- When discussing joint training as preferable to staged approaches when retrieval and reasoning have complementary supervision signals.
- When citing evidence that KG path scoring can be formulated as a sequence task compatible with PLM fine-tuning.
- When designing systems where multi-hop reasoning benefits from retrieving full reasoning chains rather than individual KG triples.