Improving Commonsense Question Answering by Graph-based Iterative Retrieval over Multiple Knowledge Sources

• DOI: 10.18653/v1/2020.coling-main.232
• Publication Date: 2021-01-08T13:58:31Z
• Authors: Qianglong Chen; Feng Ji; Haiqing Chen; Yin Zhang
• Keywords: Commonsense knowledge; Knowledge graph; Knowledge retrieval; ENCODE

In order to facilitate natural language understanding, the key is engage commonsense or background knowledge. However, how effectively in question answering systems still under exploration both research academia and industry. this paper, we propose a novel question-answering method by integrating multiple knowledge sources, i.e. ConceptNet, Wikipedia, Cambridge Dictionary, boost performance. More concretely, first introduce graph-based iterative retrieval module, which iteratively retrieves concepts entities related given its choices from sources. Afterward, use pre-trained model encode question, retrieved choices, an answer choice-aware attention mechanism fuse all hidden representations of previous modules. Finally, linear classifier for specific tasks used predict answer. Experimental results on CommonsenseQA dataset show that our significantly outperforms other competitive methods achieves new state-of-the-art. addition, further ablation studies demonstrate effectiveness module retrieving synthesizing