Dense Retrieval of Knowledge Graphs for Question Answering
- https://knowledge-nlp.github.io/kdd2023/papers/Nangi9.pdf
- Sharmila Reddy Nangi, Michihiro Yasunaga, Hongyu Ren, Qian Huang, Percy Liang, Jure Leskovec
Improves QA-GNN by Yasunaga2021QA-GNN, in particular the subgraph retrieval approach.
DrKG (dense retrieval of knowledge graph) method. It represents KG triplets as text passages and trains a dense retriever to find the relevant triplets based on the context. It then creates a subgraph from the triplets.
(this can probably done with the zipping method? or simply LangChain style look up? Can we also do edge-based look ups (quick node look up -> edge)?)