{"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/a-deep-cascade-model-for-multi-document","title":"A Deep Cascade Model for Multi-Document Reading Comprehension","arxiv_id":"1811.11374","date":"2018-11-28","proceeding":null,"authors":["Ming Yan","Jiangnan Xia","Chen Wu","Bin Bi","Zhongzhou Zhao","Ji Zhang","Luo Si","Rui Wang","Wei Wang","Haiqing Chen"],"abstract":"A fundamental trade-off between effectiveness and efficiency needs to be\nbalanced when designing an online question answering system. Effectiveness\ncomes from sophisticated functions such as extractive machine reading\ncomprehension (MRC), while efficiency is obtained from improvements in\npreliminary retrieval components such as candidate document selection and\nparagraph ranking. Given the complexity of the real-world multi-document MRC\nscenario, it is difficult to jointly optimize both in an end-to-end system. To\naddress this problem, we develop a novel deep cascade learning model, which\nprogressively evolves from the document-level and paragraph-level ranking of\ncandidate texts to more precise answer extraction with machine reading\ncomprehension. Specifically, irrelevant documents and paragraphs are first\nfiltered out with simple functions for efficiency consideration. Then we\njointly train three modules on the remaining texts for better tracking the\nanswer: the document extraction, the paragraph extraction and the answer\nextraction. Experiment results show that the proposed method outperforms the\nprevious state-of-the-art methods on two large-scale multi-document benchmark\ndatasets, i.e., TriviaQA and DuReader. In addition, our online system can\nstably serve typical scenarios with millions of daily requests in less than\n50ms.","url_abs":"http://arxiv.org/abs/1811.11374v1","url_pdf":"http://arxiv.org/pdf/1811.11374v1.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":"machine-reading-comprehension","task_name":"Machine Reading Comprehension"},{"task_slug":"question-answering","task_name":"Question Answering"},{"task_slug":"reading-comprehension","task_name":"Reading Comprehension"},{"task_slug":"retrieval","task_name":"Retrieval"},{"task_slug":"triviaqa","task_name":"TriviaQA"}],"methods":[],"datasets_introduced":[],"methods_introduced":[],"results":[{"leaderboard":"/sota/question-answering-on-ms-marco","task":"Question Answering","dataset":"MS MARCO","model":"Deep Cascade QA","rank_in_archive_order":2,"of":4,"metrics":{"BLEU-1":"54.64","Rouge-L":"52.01"},"uses_additional_data":false}],"syntology":{"atlas_url":"https://app.syntology.ai/?focus=1811.11374","mcp":null,"developers":"https://syntology.ai/developers"},"arxiv_metadata":null,"syntology_extracted_results":null}