{"about":{"site":"https://codewithpapers.app","non_affiliation":"Code with Papers and Syntology are not affiliated with, endorsed by, or sponsored by Papers with Code, Meta, or the pwc-archive mirror.","licence":"CC BY-SA 4.0","licence_url":"https://creativecommons.org/licenses/by-sa/4.0/legalcode","attribution":"https://codewithpapers.app/attribution","modified":"archive material modified by Syntology; see the attribution page"},"url":"/paper/pullnet-open-domain-question-answering-with","title":"PullNet: Open Domain Question Answering with Iterative Retrieval on Knowledge Bases and Text","arxiv_id":"1904.09537","date":"2019-04-21","proceeding":"IJCNLP 2019 11","authors":["Haitian Sun","Tania Bedrax-Weiss","William W. Cohen"],"abstract":"We consider open-domain queston answering (QA) where answers are drawn from\neither a corpus, a knowledge base (KB), or a combination of both of these. We\nfocus on a setting in which a corpus is supplemented with a large but\nincomplete KB, and on questions that require non-trivial (e.g., ``multi-hop'')\nreasoning. We describe PullNet, an integrated framework for (1) learning what\nto retrieve (from the KB and/or corpus) and (2) reasoning with this\nheterogeneous information to find the best answer. PullNet uses an {iterative}\nprocess to construct a question-specific subgraph that contains information\nrelevant to the question. In each iteration, a graph convolutional network\n(graph CNN) is used to identify subgraph nodes that should be expanded using\nretrieval (or ``pull'') operations on the corpus and/or KB. After the subgraph\nis complete, a similar graph CNN is used to extract the answer from the\nsubgraph. This retrieve-and-reason process allows us to answer multi-hop\nquestions using large KBs and corpora. PullNet is weakly supervised, requiring\nquestion-answer pairs but not gold inference paths. Experimentally PullNet\nimproves over the prior state-of-the art, and in the setting where a corpus is\nused with incomplete KB these improvements are often dramatic. PullNet is also\noften superior to prior systems in a KB-only setting or a text-only setting.","url_abs":"http://arxiv.org/abs/1904.09537v1","url_pdf":"http://arxiv.org/pdf/1904.09537v1.pdf","source":{"archive":"pwc-archive (Hugging Face), CC BY-SA 4.0","snapshot":"2025-07-28","licence_url":"https://creativecommons.org/licenses/by-sa/4.0/legalcode","row_kind":"abstracts"},"code_links":[],"tasks":[{"task_slug":"open-domain-question-answering","task_name":"Open-Domain Question Answering"},{"task_slug":"question-answering","task_name":"Question Answering"},{"task_slug":"retrieval","task_name":"Retrieval"}],"methods":[],"datasets_introduced":[],"methods_introduced":[],"results":[{"leaderboard":"/sota/question-answering-on-timequestions","task":"Question Answering","dataset":"TimeQuestions","model":"PullNet","rank_in_archive_order":21,"of":21,"metrics":{"P@1":"10.5"},"uses_additional_data":false}],"syntology":{"syntology_url":"https://syntology.ai/paper/1904.09537","atlas_url":"https://app.syntology.ai/?focus=1904.09537","mcp":null,"developers":"https://syntology.ai/developers"},"arxiv_metadata":null,"syntology_extracted_results":null}