{"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/reasonet-learning-to-stop-reading-in-machine","title":"ReasoNet: Learning to Stop Reading in Machine Comprehension","arxiv_id":"1609.05284","date":"2016-09-17","proceeding":null,"authors":["Yelong Shen","Po-Sen Huang","Jianfeng Gao","Weizhu Chen"],"abstract":"Teaching a computer to read and answer general questions pertaining to a\ndocument is a challenging yet unsolved problem. In this paper, we describe a\nnovel neural network architecture called the Reasoning Network (ReasoNet) for\nmachine comprehension tasks. ReasoNets make use of multiple turns to\neffectively exploit and then reason over the relation among queries, documents,\nand answers. Different from previous approaches using a fixed number of turns\nduring inference, ReasoNets introduce a termination state to relax this\nconstraint on the reasoning depth. With the use of reinforcement learning,\nReasoNets can dynamically determine whether to continue the comprehension\nprocess after digesting intermediate results, or to terminate reading when it\nconcludes that existing information is adequate to produce an answer. ReasoNets\nhave achieved exceptional performance in machine comprehension datasets,\nincluding unstructured CNN and Daily Mail datasets, the Stanford SQuAD dataset,\nand a structured Graph Reachability dataset.","url_abs":"http://arxiv.org/abs/1609.05284v3","url_pdf":"http://arxiv.org/pdf/1609.05284v3.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":"question-answering","task_name":"Question Answering"},{"task_slug":"reading-comprehension","task_name":"Reading Comprehension"},{"task_slug":"reinforcement-learning","task_name":"Reinforcement Learning"}],"methods":[],"datasets_introduced":[],"methods_introduced":[],"results":[{"leaderboard":"/sota/question-answering-on-cnn-daily-mail","task":"Question Answering","dataset":"CNN / Daily Mail","model":"ReasoNet","rank_in_archive_order":7,"of":16,"metrics":{"CNN":"74.7","Daily Mail":"76.6"},"uses_additional_data":false},{"leaderboard":"/sota/question-answering-on-squad11","task":"Question Answering","dataset":"SQuAD1.1","model":"ReasoNet (ensemble)","rank_in_archive_order":123,"of":213,"metrics":{"EM":"75.034","F1":"82.552"},"uses_additional_data":true},{"leaderboard":"/sota/question-answering-on-squad11","task":"Question Answering","dataset":"SQuAD1.1","model":"ReasoNet (single model)","rank_in_archive_order":161,"of":213,"metrics":{"EM":"70.555","F1":"79.364"},"uses_additional_data":false}],"syntology":{"atlas_url":"https://app.syntology.ai/?focus=1609.05284","mcp":null,"developers":"https://syntology.ai/developers"},"arxiv_metadata":null,"syntology_extracted_results":null}