CBR-iKB: A Case-Based Reasoning Approach for Question Answering over Incomplete Knowledge Bases
Dung Thai,Srinivas Ravishankar,Ibrahim Abdelaziz,Mudit Chaudhary,Nandana Mihindukulasooriya,Tahira Naseem,Rajarshi Das,Pavan Kapanipathi,Achille Fokoue,Andrew McCallum
DOI: https://doi.org/10.48550/arXiv.2204.08554
2022-04-19
Abstract:Knowledge bases (KBs) are often incomplete and constantly changing in practice. Yet, in many question answering applications coupled with knowledge bases, the sparse nature of KBs is often overlooked. To this end, we propose a case-based reasoning approach, CBR-iKB, for knowledge base question answering (KBQA) with incomplete-KB as our main focus. Our method ensembles decisions from multiple reasoning chains with a novel nonparametric reasoning algorithm. By design, CBR-iKB can seamlessly adapt to changes in KBs without any task-specific training or fine-tuning. Our method achieves 100% accuracy on MetaQA and establishes new state-of-the-art on multiple benchmarks. For instance, CBR-iKB achieves an accuracy of 70% on WebQSP under the incomplete-KB setting, outperforming the existing state-of-the-art method by 22.3%.
Computation and Language,Artificial Intelligence,Information Retrieval,Machine Learning