{"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/multi-reward-reinforced-summarization-with","title":"Multi-Reward Reinforced Summarization with Saliency and Entailment","arxiv_id":"1804.06451","date":"2018-04-17","proceeding":"NAACL 2018 6","authors":["Ramakanth Pasunuru","Mohit Bansal"],"abstract":"Abstractive text summarization is the task of compressing and rewriting a\nlong document into a short summary while maintaining saliency, directed logical\nentailment, and non-redundancy. In this work, we address these three important\naspects of a good summary via a reinforcement learning approach with two novel\nreward functions: ROUGESal and Entail, on top of a coverage-based baseline. The\nROUGESal reward modifies the ROUGE metric by up-weighting the salient\nphrases/words detected via a keyphrase classifier. The Entail reward gives high\n(length-normalized) scores to logically-entailed summaries using an entailment\nclassifier. Further, we show superior performance improvement when these\nrewards are combined with traditional metric (ROUGE) based rewards, via our\nnovel and effective multi-reward approach of optimizing multiple rewards\nsimultaneously in alternate mini-batches. Our method achieves the new\nstate-of-the-art results (including human evaluation) on the CNN/Daily Mail\ndataset as well as strong improvements in a test-only transfer setup on\nDUC-2002.","url_abs":"http://arxiv.org/abs/1804.06451v2","url_pdf":"http://arxiv.org/pdf/1804.06451v2.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":"abstractive-text-summarization","task_name":"Abstractive Text Summarization"},{"task_slug":"reinforcement-learning","task_name":"Reinforcement Learning"},{"task_slug":"text-summarization","task_name":"Text Summarization"}],"methods":[],"datasets_introduced":[],"methods_introduced":[],"results":[{"leaderboard":"/sota/abstractive-text-summarization-on-cnn-daily","task":"Abstractive Text Summarization","dataset":"CNN / Daily Mail","model":"ROUGESal+Ent RL","rank_in_archive_order":41,"of":53,"metrics":{"ROUGE-1":"40.43","ROUGE-2":"18.00","ROUGE-L":"37.10"},"uses_additional_data":false}],"syntology":{"atlas_url":"https://app.syntology.ai/?focus=1804.06451","mcp":null,"developers":"https://syntology.ai/developers"},"arxiv_metadata":null,"syntology_extracted_results":null}