| { | |
| "paper_id": "2019", | |
| "header": { | |
| "generated_with": "S2ORC 1.0.0", | |
| "date_generated": "2023-01-19T14:54:45.435746Z" | |
| }, | |
| "title": "GALs: \u57fa\u65bc\u5c0d\u6297\u5f0f\u5b78\u7fd2\u4e4b\u6574\u5217\u5f0f\u6458\u8981\u6cd5 GALs: A GAN-based Listwise Summarizer", | |
| "authors": [ | |
| { | |
| "first": "Chia-Chih", | |
| "middle": [], | |
| "last": "\u90ed\u5bb6\u928d", | |
| "suffix": "", | |
| "affiliation": { | |
| "laboratory": "", | |
| "institution": "National Taiwan University of Science and Technology", | |
| "location": {} | |
| }, | |
| "email": "" | |
| }, | |
| { | |
| "first": "", | |
| "middle": [], | |
| "last": "Kuo", | |
| "suffix": "", | |
| "affiliation": { | |
| "laboratory": "", | |
| "institution": "National Taiwan University of Science and Technology", | |
| "location": {} | |
| }, | |
| "email": "" | |
| }, | |
| { | |
| "first": "Kuan-Yu", | |
| "middle": [], | |
| "last": "\u9673\u51a0\u5b87", | |
| "suffix": "", | |
| "affiliation": { | |
| "laboratory": "", | |
| "institution": "National Taiwan University of Science and Technology", | |
| "location": {} | |
| }, | |
| "email": "" | |
| }, | |
| { | |
| "first": "", | |
| "middle": [], | |
| "last": "Chen", | |
| "suffix": "", | |
| "affiliation": { | |
| "laboratory": "", | |
| "institution": "National Taiwan University of Science and Technology", | |
| "location": {} | |
| }, | |
| "email": "[email protected]" | |
| } | |
| ], | |
| "year": "", | |
| "venue": null, | |
| "identifiers": {}, | |
| "abstract": "Extractive summarization aims at selecting a set of sentences to form a summary for a given document. Learning-to-rank is first appeared in the field of information retrieval, and it has been employed to solve several ranking-based tasks. In this study, we regard the task of extractive summarization as a listwise sentence ranking problem, and thus a GAN-based", | |
| "pdf_parse": { | |
| "paper_id": "2019", | |
| "_pdf_hash": "", | |
| "abstract": [ | |
| { | |
| "text": "Extractive summarization aims at selecting a set of sentences to form a summary for a given document. Learning-to-rank is first appeared in the field of information retrieval, and it has been employed to solve several ranking-based tasks. In this study, we regard the task of extractive summarization as a listwise sentence ranking problem, and thus a GAN-based", | |
| "cite_spans": [], | |
| "ref_spans": [], | |
| "eq_spans": [], | |
| "section": "Abstract", | |
| "sec_num": null | |
| } | |
| ], | |
| "body_text": [ | |
| { | |
| "text": "listwise summarizer (GALs) is proposed. On top of the generative adversarial network (GAN), an extractive summarizer is introduced to be the generator, and a discriminator is employed to distinguish the generated summary from the ground truth. Especially, the input to the discriminator is a set of surface features, which are extracted from the generated summary and the ground truth. Finally, GALs can be optimized by leveraging the reinforcement learning (RL) strategy. The experimental results demonstrate the effectiveness of the proposed framework on the CNN/Daily Mail corpus. Moreover, we make detailed investigation and analysis of the parameters used in GALs. ", | |
| "cite_spans": [], | |
| "ref_spans": [], | |
| "eq_spans": [], | |
| "section": "", | |
| "sec_num": null | |
| }, | |
| { | |
| "text": "EQUATION", | |
| "cite_spans": [], | |
| "ref_spans": [], | |
| "eq_spans": [ | |
| { | |
| "start": 0, | |
| "end": 8, | |
| "text": "EQUATION", | |
| "ref_id": "EQREF", | |
| "raw_str": "= softmax( \u22a4 tanh\ufffd 1 \u210e + 2 \ufffd (1) = \u2211 1 \u210e (2) \u63a5\u8457\u5c07\u524d\u5f8c\u6587\u5411\u91cf \u518d\u6b21\u8207\u6587\u672c\u4e2d\u7684\u53e5\u5b50\u5411\u91cf\u210e \u8a08\u7b97\u6ce8\u610f\u529b\uff0c\u4e26\u6392\u9664\u904e\u53bb\u7684\u6642\u9593\u9ede\u4e2d \u5df2\u62bd\u53d6\u7684\u53e5\u5b50 1 , \u2026 , \u22121 \uff0c\u6700\u5f8c\u53d6\u5f97\u7576\u524d\u6642\u9593\u9ede\u62bd\u53d6\u53e5\u5b50\u4e4b\u6a5f\u7387\u5206\u4f48 ( | 1 , \u2026 , \u22121 )\u3002 \u5716\u4e00\u3001\u57fa\u65bc\u5c0d\u6297\u5f0f\u5b78\u7fd2\u4e4b\u6574\u5217\u5f0f\u6458\u8981\u6cd5(GALs)\u4e4b\u6a21\u578b\u8a13\u7df4\u67b6\u69cb = \ufffd \u22a4 tanh\ufffd 1 \u210e + 2 \ufffd, if \u2260 \u2200 < \u2212\u221e, otherwise", | |
| "eq_num": "(3)" | |
| } | |
| ], | |
| "section": "", | |
| "sec_num": null | |
| }, | |
| { | |
| "text": "EQUATION", | |
| "cite_spans": [], | |
| "ref_spans": [], | |
| "eq_spans": [ | |
| { | |
| "start": 0, | |
| "end": 8, | |
| "text": "EQUATION", | |
| "ref_id": "EQREF", | |
| "raw_str": "( | 1 , \u2026 , \u22121 ) = softmax( )", | |
| "eq_num": "(4)" | |
| } | |
| ], | |
| "section": "", | |
| "sec_num": null | |
| } | |
| ], | |
| "back_matter": [], | |
| "bib_entries": { | |
| "BIBREF0": { | |
| "ref_id": "b0", | |
| "title": "Distributed representations of words and phrases and their compositionality", | |
| "authors": [ | |
| { | |
| "first": "T", | |
| "middle": [], | |
| "last": "Mikolov", | |
| "suffix": "" | |
| }, | |
| { | |
| "first": "J", | |
| "middle": [], | |
| "last": "Dean", | |
| "suffix": "" | |
| } | |
| ], | |
| "year": 2013, | |
| "venue": "Proceedings of the 26th International Conference on Neural Information Processing Systems", | |
| "volume": "2", | |
| "issue": "", | |
| "pages": "3111--3119", | |
| "other_ids": {}, | |
| "num": null, | |
| "urls": [], | |
| "raw_text": "T. Mikolov, J. Dean, \"Distributed representations of words and phrases and their compositionality,\" Proceedings of the 26th International Conference on Neural Information Processing Systems, vol. 2, pp. 3111-3119, 2013.", | |
| "links": null | |
| }, | |
| "BIBREF1": { | |
| "ref_id": "b1", | |
| "title": "Long Short-Term Memory", | |
| "authors": [ | |
| { | |
| "first": "S", | |
| "middle": [], | |
| "last": "Hochreiter", | |
| "suffix": "" | |
| }, | |
| { | |
| "first": "J", | |
| "middle": [], | |
| "last": "Schmidhuber", | |
| "suffix": "" | |
| } | |
| ], | |
| "year": 1997, | |
| "venue": "Neural Computation", | |
| "volume": "9", | |
| "issue": "8", | |
| "pages": "1735--1780", | |
| "other_ids": {}, | |
| "num": null, | |
| "urls": [], | |
| "raw_text": "S. Hochreiter, J. Schmidhuber, \"Long Short-Term Memory,\" Neural Computation, vol. 9, no. 8, pp. 1735-1780, 1997.", | |
| "links": null | |
| }, | |
| "BIBREF2": { | |
| "ref_id": "b2", | |
| "title": "IRGAN: A minimax game for unifying generative and discriminative information retrieval models", | |
| "authors": [ | |
| { | |
| "first": "J", | |
| "middle": [], | |
| "last": "Wang", | |
| "suffix": "" | |
| }, | |
| { | |
| "first": "L", | |
| "middle": [], | |
| "last": "Yu", | |
| "suffix": "" | |
| }, | |
| { | |
| "first": "W", | |
| "middle": [], | |
| "last": "Zhang", | |
| "suffix": "" | |
| }, | |
| { | |
| "first": "Y", | |
| "middle": [], | |
| "last": "Gong", | |
| "suffix": "" | |
| }, | |
| { | |
| "first": "Y", | |
| "middle": [], | |
| "last": "Xu", | |
| "suffix": "" | |
| }, | |
| { | |
| "first": "B", | |
| "middle": [], | |
| "last": "Wang", | |
| "suffix": "" | |
| }, | |
| { | |
| "first": "P", | |
| "middle": [], | |
| "last": "Zhang", | |
| "suffix": "" | |
| }, | |
| { | |
| "first": "D", | |
| "middle": [], | |
| "last": "Zhang", | |
| "suffix": "" | |
| } | |
| ], | |
| "year": 2017, | |
| "venue": "Proceedings of the 40th International ACM SIGIR Conference on Research and Development in Information Retrieval", | |
| "volume": "", | |
| "issue": "", | |
| "pages": "515--524", | |
| "other_ids": {}, | |
| "num": null, | |
| "urls": [], | |
| "raw_text": "J. Wang, L. Yu, W. Zhang, Y. Gong, Y. Xu, B. Wang, P. Zhang, D. Zhang, \"IRGAN: A minimax game for unifying generative and discriminative information retrieval models,\" Proceedings of the 40th International ACM SIGIR Conference on Research and Development in Information Retrieval, pp. 515-524, 2017.", | |
| "links": null | |
| }, | |
| "BIBREF3": { | |
| "ref_id": "b3", | |
| "title": "ROUGE: A package for automatic evaluation of summaries", | |
| "authors": [ | |
| { | |
| "first": "C.-Y.", | |
| "middle": [], | |
| "last": "Lin", | |
| "suffix": "" | |
| } | |
| ], | |
| "year": 2004, | |
| "venue": "", | |
| "volume": "", | |
| "issue": "", | |
| "pages": "74--81", | |
| "other_ids": {}, | |
| "num": null, | |
| "urls": [], | |
| "raw_text": "C.-Y. Lin, \"ROUGE: A package for automatic evaluation of summaries,\" Text Summarization Branches Out, pp. 74-81, 2004.", | |
| "links": null | |
| }, | |
| "BIBREF4": { | |
| "ref_id": "b4", | |
| "title": "Sequence to sequence learning with neural networks", | |
| "authors": [ | |
| { | |
| "first": "I", | |
| "middle": [], | |
| "last": "Sutskever", | |
| "suffix": "" | |
| }, | |
| { | |
| "first": "O", | |
| "middle": [], | |
| "last": "Vinyals", | |
| "suffix": "" | |
| }, | |
| { | |
| "first": "Q", | |
| "middle": [ | |
| "V" | |
| ], | |
| "last": "Le", | |
| "suffix": "" | |
| } | |
| ], | |
| "year": 2014, | |
| "venue": "Proceedings of the 27th International Conference on Neural Information Processing Systems", | |
| "volume": "2", | |
| "issue": "", | |
| "pages": "3104--3112", | |
| "other_ids": {}, | |
| "num": null, | |
| "urls": [], | |
| "raw_text": "I. Sutskever, O. Vinyals, Q. V. Le, \"Sequence to sequence learning with neural networks,\" Proceedings of the 27th International Conference on Neural Information Processing Systems, vol. 2, pp. 3104-3112, 2014.", | |
| "links": null | |
| }, | |
| "BIBREF5": { | |
| "ref_id": "b5", | |
| "title": "Neural machine translation by jointly learning to align and translate", | |
| "authors": [ | |
| { | |
| "first": "D", | |
| "middle": [], | |
| "last": "Bahdanau", | |
| "suffix": "" | |
| }, | |
| { | |
| "first": "K", | |
| "middle": [], | |
| "last": "Cho", | |
| "suffix": "" | |
| }, | |
| { | |
| "first": "Y", | |
| "middle": [], | |
| "last": "Bengio", | |
| "suffix": "" | |
| } | |
| ], | |
| "year": 2015, | |
| "venue": "International Conference on Learning Representations", | |
| "volume": "", | |
| "issue": "", | |
| "pages": "", | |
| "other_ids": {}, | |
| "num": null, | |
| "urls": [], | |
| "raw_text": "D. Bahdanau, K. Cho, Y. Bengio, \"Neural machine translation by jointly learning to align and translate,\" International Conference on Learning Representations, 2015.", | |
| "links": null | |
| }, | |
| "BIBREF6": { | |
| "ref_id": "b6", | |
| "title": "The multilayer perceptron as an approximation to a Bayes optimal discriminant function", | |
| "authors": [ | |
| { | |
| "first": "D", | |
| "middle": [ | |
| "W" | |
| ], | |
| "last": "Ruck", | |
| "suffix": "" | |
| }, | |
| { | |
| "first": "S", | |
| "middle": [ | |
| "K" | |
| ], | |
| "last": "Rogers", | |
| "suffix": "" | |
| }, | |
| { | |
| "first": "M", | |
| "middle": [], | |
| "last": "Kabrisky", | |
| "suffix": "" | |
| }, | |
| { | |
| "first": "M", | |
| "middle": [ | |
| "E" | |
| ], | |
| "last": "Oxley", | |
| "suffix": "" | |
| }, | |
| { | |
| "first": "B", | |
| "middle": [ | |
| "W" | |
| ], | |
| "last": "Suter", | |
| "suffix": "" | |
| } | |
| ], | |
| "year": 1990, | |
| "venue": "IEEE Transactions on Neural Networks", | |
| "volume": "1", | |
| "issue": "4", | |
| "pages": "296--298", | |
| "other_ids": {}, | |
| "num": null, | |
| "urls": [], | |
| "raw_text": "D. W. Ruck, S. K. Rogers, M. Kabrisky, M. E. Oxley, B. W. Suter, \"The multilayer perceptron as an approximation to a Bayes optimal discriminant function,\" IEEE Transactions on Neural Networks, vol. 1, no. 4, pp. 296-298, 1990.", | |
| "links": null | |
| }, | |
| "BIBREF7": { | |
| "ref_id": "b7", | |
| "title": "Gene selection for cancer classification using support vector machines", | |
| "authors": [ | |
| { | |
| "first": "I", | |
| "middle": [], | |
| "last": "Guyon", | |
| "suffix": "" | |
| }, | |
| { | |
| "first": "J", | |
| "middle": [], | |
| "last": "Weston", | |
| "suffix": "" | |
| }, | |
| { | |
| "first": "S", | |
| "middle": [], | |
| "last": "Barnhill", | |
| "suffix": "" | |
| }, | |
| { | |
| "first": "V", | |
| "middle": [ | |
| "N" | |
| ], | |
| "last": "Vapnik", | |
| "suffix": "" | |
| } | |
| ], | |
| "year": 2002, | |
| "venue": "Machine Learning", | |
| "volume": "46", | |
| "issue": "", | |
| "pages": "389--422", | |
| "other_ids": {}, | |
| "num": null, | |
| "urls": [], | |
| "raw_text": "I. Guyon, J. Weston, S. Barnhill, V. N. Vapnik, \"Gene selection for cancer classification using support vector machines,\" Machine Learning, vol. 46, no. 13, pp. 389-422, 2002.", | |
| "links": null | |
| }, | |
| "BIBREF8": { | |
| "ref_id": "b8", | |
| "title": "Unsupervised learning of sentence embeddings using compositional n-gram features", | |
| "authors": [ | |
| { | |
| "first": "M", | |
| "middle": [], | |
| "last": "Pagliardini", | |
| "suffix": "" | |
| }, | |
| { | |
| "first": "P", | |
| "middle": [], | |
| "last": "Gupta", | |
| "suffix": "" | |
| }, | |
| { | |
| "first": "M", | |
| "middle": [], | |
| "last": "Jaggi", | |
| "suffix": "" | |
| } | |
| ], | |
| "year": 2018, | |
| "venue": "Proceedings of the 2018 Conference of the North American Chapter of the Association for Computational Linguistics: Human Language Technologies", | |
| "volume": "1", | |
| "issue": "", | |
| "pages": "528--540", | |
| "other_ids": {}, | |
| "num": null, | |
| "urls": [], | |
| "raw_text": "M. Pagliardini, P. Gupta, M. Jaggi, \"Unsupervised learning of sentence embeddings using compositional n-gram features,\" Proceedings of the 2018 Conference of the North American Chapter of the Association for Computational Linguistics: Human Language Technologies, vol. 1, pp. 528-540, 2018.", | |
| "links": null | |
| }, | |
| "BIBREF9": { | |
| "ref_id": "b9", | |
| "title": "Ranking sentences for extractive summarization with reinforcement learning", | |
| "authors": [ | |
| { | |
| "first": "S", | |
| "middle": [], | |
| "last": "Narayan", | |
| "suffix": "" | |
| }, | |
| { | |
| "first": "S", | |
| "middle": [ | |
| "B" | |
| ], | |
| "last": "Cohen", | |
| "suffix": "" | |
| }, | |
| { | |
| "first": "M", | |
| "middle": [], | |
| "last": "Lapata", | |
| "suffix": "" | |
| } | |
| ], | |
| "year": 2018, | |
| "venue": "Proceedings of the 2018 Conference of the North American Chapter of the Association for Computational Linguistics: Human Language Technologies", | |
| "volume": "1", | |
| "issue": "", | |
| "pages": "1747--1759", | |
| "other_ids": {}, | |
| "num": null, | |
| "urls": [], | |
| "raw_text": "S. Narayan, S. B. Cohen, M. Lapata, \"Ranking sentences for extractive summarization with reinforcement learning,\" Proceedings of the 2018 Conference of the North American Chapter of the Association for Computational Linguistics: Human Language Technologies, vol. 1, pp. 1747-1759, 2018.", | |
| "links": null | |
| }, | |
| "BIBREF10": { | |
| "ref_id": "b10", | |
| "title": "Neural latent extractive document summarization", | |
| "authors": [ | |
| { | |
| "first": "X", | |
| "middle": [], | |
| "last": "Zhang", | |
| "suffix": "" | |
| }, | |
| { | |
| "first": "M", | |
| "middle": [], | |
| "last": "Lapata", | |
| "suffix": "" | |
| }, | |
| { | |
| "first": "F", | |
| "middle": [], | |
| "last": "Wei", | |
| "suffix": "" | |
| }, | |
| { | |
| "first": "M", | |
| "middle": [], | |
| "last": "Zhou", | |
| "suffix": "" | |
| } | |
| ], | |
| "year": 2018, | |
| "venue": "Proceedings of the 2018 Conference on Empirical Methods in Natural Language Processing", | |
| "volume": "", | |
| "issue": "", | |
| "pages": "779--784", | |
| "other_ids": {}, | |
| "num": null, | |
| "urls": [], | |
| "raw_text": "X. Zhang, M. Lapata, F. Wei, M. Zhou, \"Neural latent extractive document summarization,\" Proceedings of the 2018 Conference on Empirical Methods in Natural Language Processing, pp. 779-784, 2018.", | |
| "links": null | |
| } | |
| }, | |
| "ref_entries": { | |
| "TABREF1": { | |
| "num": null, | |
| "type_str": "table", | |
| "content": "<table><tr><td>\u57fa\u65bc\u4e0a\u8ff0\u89e3\u78bc\u6d41\u7a0b\uff0c\u89e3\u78bc\u5668\u5c07\u905e\u56de\u62bd\u53d6\u53e5\u5b50\uff0c\u76f4\u5230\u89f8\u767c\u8a2d\u5b9a\u4e4b\u505c\u6b62\u689d\u4ef6\u3002 (\u4e09)\u5224\u5225\u5668 GALs \u7684\u5224\u5225\u5668\u662f\u4e00\u500b\u7528\u65bc\u5206\u8fa8\u6b63\u8ca0\u6a23\u672c\u7684\u4e8c\u5143\u5206\u985e\u5668\uff0c\u6211\u5011\u55ae\u7d14\u63a1\u7528\u591a\u5c64\u611f\u77e5\u5668 (Multilayer Perceptron)[11]\u7d44\u6210\u4e4b\u985e\u795e\u7d93\u7db2\u8def\u6a21\u578b\u4f5c\u70ba\u5224\u5225\u5668\u3002\u56e0\u70ba\u4e00\u4efd\u6458\u8981\u901a\u5e38\u5305 \u542b\u4e86\u591a\u500b\u53e5\u5b50\uff0c\u56e0\u6b64\u6211\u5011\u8a08\u7b97\u53e5\u5b50\u9593\u5404\u500b\u7279\u5fb5\u7684\u5e73\u5747\u503c\u8207\u6a19\u6e96\u5dee\u4f5c\u70ba\u5224\u5225\u5668\u7684\u8f38\u5165\uff1b \u82e5\u6458\u8981\u4e2d\u53ea\u6709\u4e00\u500b\u53e5\u5b50\uff0c\u5247\u8a2d\u5b9a\u6a19\u6e96\u5dee\u70ba\u96f6\u3002\u70ba\u4e86\u4f7f\u5224\u5225\u5668\u80fd\u63d0\u4f9b\u66f4\u5bcc\u50f9\u503c\u4e4b\u734e\u52f5\uff0c \u6211\u5011\u5229\u7528\u905e\u6b78\u7279\u5fb5\u6d88\u9664(Recursive Feature Elimination, RFE)[12] \u7be9\u9664\u4e0d\u5177\u6548\u76ca\u4e4b\u7279 \u5716\u4e8c\u3001\u57fa\u65bc\u6307\u91dd\u7db2\u8def\u4e4b\u62bd\u53d6\u5f0f\u6458\u8981\u5668 3\u3001\u76f8\u5c0d\u4f4d\u7f6e \u64b0\u7b46\u4eba\u5e38\u5e38\u6703\u5c07\u591a\u500b\u91cd\u8981\u7684\u53e5\u5b50\u96c6\u4e2d\u5728\u6587\u672c\u4e2d\u7684\u4e00\u8655\uff0c\u56e0\u6b64\u6211\u5011\u5c07\u53e5\u5b50\u7684\u7d55\u5c0d\u4f4d\u7f6e\u9664 \u4ee5\u6587\u672c\u7684\u7e3d\u53e5\u6578\uff0c\u8a08\u7b97\u5f97\u76f8\u5c0d\u4f4d\u7f6e\u4f5c\u70ba\u7279\u5fb5\u3002 4\u3001\u505c\u7528\u8a5e\u6bd4\u4f8b \u505c\u7528\u8a5e(Stop Words)\u610f\u6307\u8a9e\u8a00\u4e2d\u6975\u5e38\u51fa\u73fe\u7684\u8a5e\u5f59(\u4f8b\u5982\u82f1\u6587\u4e2d\u7684 to, the, and \u7b49\u7b49)\uff0c \u5177\u6709\u8f03\u9ad8\u505c\u7528\u8a5e\u6bd4\u4f8b\u4e4b\u53e5\u5b50\u5f88\u6709\u53ef\u80fd\u6bd4\u5176\u4ed6\u53e5\u5b50\u4f86\u5f97\u66f4\u4e0d\u5177\u5be6\u969b\u610f\u7fa9\u3002 5\u3001\u5e73\u5747\u8a5e\u983b \u8a5e\u983b(Word Frequency)\u5373\u8a5e\u5728\u6587\u672c\u4e2d\u51fa\u73fe\u7684\u983b\u7387\uff0c\u53ef\u7528\u65bc\u8a55\u4f30\u8a5e\u5728\u6587\u672c\u4e2d\u7684\u91cd\u8981 \u6027\u3002\u6211\u5011\u8a08\u7b97\u53e5\u5b50\u4e2d\u6240\u6709\u8a5e\u4e4b\u8a5e\u983b\u7684\u5e73\u5747\u503c\u4f5c\uff0c\u5e73\u5747\u8a5e\u983b\u8f03\u9ad8\u7684\u53e5\u5b50\u53ef\u80fd\u66f4\u70ba\u91cd\u8981\u3002 6\u3001\u5e73\u5747\u6587\u672c\u983b \u6587\u672c\u983b(Document Frequency)\u5e38\u8207\u8a5e\u983b\u4e00\u8d77\u4f7f\u7528\uff0c\u82e5\u67d0\u4e00\u8a5e\u5f59\u5728\u8a9e\u6599\u4e2d\u7684\u5927\u591a\u6578\u6587\u672c \u4e2d\u90fd\u51fa\u73fe\u904e(\u5373\u6587\u672c\u983b\u8f03\u9ad8)\uff0c\u5247\u6b64\u8a5e\u5f59\u96e3\u4ee5\u7a81\u986f\u51fa\u53e5\u5b50\u9593\u7684\u5dee\u7570\u6027\u3002\u6211\u5011\u8a08\u7b97\u53e5\u5b50 \u4e2d\u6240\u6709\u8a5e\u5f59\u7684\u6587\u672c\u983b\u5e73\u5747\u503c\u4f5c\u70ba\u7279\u5fb5\uff0c\u5e73\u5747\u6587\u672c\u983b\u8f03\u9ad8\u7684\u53e5\u5b50\u53ef\u80fd\u8f03\u4e0d\u91cd\u8981\u3002 7\u3001\u53e5\u5d4c\u5165 \u6211\u5011\u4f7f\u7528 Sent2Vec [13]\uff0c\u4e00\u7a2e\u65e8\u65bc\u63a2\u7a76\u8a9e\u610f\u7279\u5fb5\u7684\u975e\u76e3\u7763\u65b9\u6cd5\uff0c\u7528\u4ee5\u5c07\u53e5\u5b50\u5f62\u6210\u5206\u6563 \u5f0f\u8868\u793a\u6cd5(Distributed Representations)\uff0c\u4e5f\u7d0d\u5165\u8003\u91cf\u4e4b\u4e2d\u3002\u6211\u5011\u65bc CNN/Daily Mail \u7684 \u8a13\u7df4\u96c6\u4e0a\u8a13\u7df4 100 \u7dad\u7684 Sent2Vec \u6a21\u578b\uff0c\u4ee5\u751f\u6210\u5404\u500b\u53e5\u5b50\u7684\u53e5\u5d4c\u5165\u5411\u91cf\u3002 8\u3001\u9918\u5f26\u76f8\u4f3c\u5ea6 \u71b5(Binary Cross-Entropy)\uff0c\u6539\u63a1\u5c07\u8ca0\u6a23\u672c\u4e4b\u9810\u6e2c\u5206\u6578\u6e1b\u53bb\u6b63\u6a23\u672c\u4e4b\u9810\u6e2c\u5206\u6578\u4f5c\u70ba\u5224 \u5225\u5668\u7684\u640d\u5931\u51fd\u6578\uff0c\u6b64\u640d\u5931\u51fd\u6578\u540c\u6642\u4ea6\u5c07\u4f5c\u70ba\u751f\u6210\u5668\u5728\u8a13\u7df4\u6642\u4f7f\u7528\u7684\u734e\u52f5\u503c\u3002\u6211\u5011\u6ce8\u610f \u5230\u56e0\u70ba\u6b64\u7a2e\u8a08\u7b97\u65b9\u5f0f\u6709\u53ef\u80fd\u7522\u751f\u8ca0\u503c\uff0c\u9019\u4f7f\u5f97\u751f\u6210\u5668\u8207\u5224\u5225\u5668\u53ef\u80fd\u50be\u5411\u65bc\u653b\u64ca\u5f7c\u6b64\u800c \u975e\u512a\u5316\u81ea\u8eab\uff0c\u56e0\u6b64\u6211\u5011\u5c07\u8ca0\u503c\u7686\u6b78\u70ba 0\uff1a = \ufffd0, \u2212 \ufffd (5) \u800c\u5728\u512a\u5316\u751f\u6210\u5668\u6642\uff0c\u6211\u5011\u9996\u5148\u4f7f\u7528\u751f\u6210\u5668\u7522\u751f\u6700\u591a\u524d\u4e94\u500b\u6642\u9593\u9ede\u7684\u9810\u6e2c\u6a5f\u7387\u5206\u4f48 \uff0c\u5206\u5225\u5f9e\u4e2d\u53d6\u6a23\u4e00\u500b\u53e5\u5b50 \uff0c\u4e26\u52a0\u7e3d\u88ab\u53d6\u6a23\u53e5\u5b50\u4e4b\u8ca0\u5c0d\u6578\u4f3c\u7136(Negative Log-likelihood)\u4e58\u4e0a\u5176\u734e\u52f5\u503c(\u5373 )\u4f5c\u70ba\u751f\u6210\u5668\u7684\u640d\u5931\u51fd\u6578\uff1a = \ufffd~( | ) [\u2212 ( | )] \u22c5 5 =1 (6) (\u4e94)\u8a13\u7df4\u6d41\u7a0b \u6211\u5011\u9996\u5148\u5c0d\u53c3\u8003\u6458\u8981\u4e2d\u7684\u6bcf\u500b\u53e5\u5b50\uff0c\u5728\u6587\u672c\u4e2d\u63d0\u53d6\u8207\u5176 ROUGE-L \u6700\u9ad8\u7684\u53e5\u5b50\u7d44\u6210\u62bd \u53d6\u5f0f\u6458\u8981\u7684\u53c3\u8003\u7b54\u6848\uff0c\u4ee5\u9810\u8a13\u7df4\u62bd\u53d6\u5f0f\u6458\u8981\u5668\u3002\u63a5\u8457\uff0c\u6211\u5011\u5c07\u9810\u8a13\u7df4\u904e\u7684\u62bd\u53d6\u5f0f\u6458\u8981 \u5668\u4f5c\u70ba GALs \u7684\u751f\u6210\u5668\uff0c\u96a8\u6a5f\u521d\u59cb\u5316\u5224\u5225\u5668\u7684\u53c3\u6578\uff0c\u4e26\u4ea4\u66ff\u8a13\u7df4\u751f\u6210\u5668\u8207\u5224\u5225\u5668\uff0c\u4ee5 \u5c07\u4f5c\u70ba\u62bd\u53d6\u5f0f\u6458\u8981\u5668\u9032\u4e00\u6b65\u512a\u5316\uff0c\u76f4\u5230\u5728\u9a57\u8b49\u96c6\u4e0a\u7372\u5f97\u6700\u4f73\u7684 ROUGE-1 \u8a55\u5206\u3002 \u56db\u3001\u5be6\u9a57 (\u4e00)CNN/Daily Mail \u6578\u64da\u96c6 ROUGE-L \u7684\u6210\u7e3e\u8868\u73fe\uff0c\u767c\u73fe\u4e09\u8005\u6536\u6582\u4e4b\u6642\u9593\u9ede\u4e0d\u540c\u3002\u70ba\u4e86\u6700\u5927\u5316 GALs \u7684\u512a\u5316\u6548\u679c\uff0c \u6211\u5011\u5617\u8a66\u5206\u5225\u4ee5\u9a57\u8b49\u96c6\u4e0a\u7684 ROUGE-1\u3001ROUGE-2 \u8207 ROUGE-L \u6307\u6a19\uff0c\u53c3\u8003\u5176\u6f32\u8dcc\u4ee5\u81ea \u52d5\u6c7a\u5b9a\u65e9\u505c\u6cd5(Early Stopping)\u8207\u5b78\u7fd2\u7387\u8870\u6e1b\u7684\u6642\u6a5f\u3002\u5728\u5be6\u9a57\u65bc CNN/Daily Mail \u6578\u64da \u96c6\u4e4b\u7d50\u679c\u986f\u793a(\u898b\u8868\u4e8c)\uff0c\u53c3\u8003 ROUGE-1 \u80fd\u5e36\u4f86\u6700\u4f73\u7684\u6574\u9ad4 ROUGE \u6210\u7e3e (\u4e09)\u512a\u5316\u7684\u53e5\u5b50\u6578\u91cf \u5728 CNN/Daily Mail \u6578\u64da\u96c6\u4e2d\uff0c\u5927\u591a\u6578\u6587\u672c\u7684\u53c3\u8003\u6458\u8981\u53ea\u5305\u542b\u4e86\u4e09\u81f3\u56db\u53e5\u8a71\uff0c\u5c0d\u6b64 Yen-Chun Chen \u8a2d\u5b9a\u62bd\u53d6\u5f0f\u6458\u8981\u5668\u56fa\u5b9a\u62bd\u53d6\u524d\u4e09\u53e5\u8a71\uff0c\u4ee5\u53d6\u5f97\u6700\u4f73\u7684\u62bd\u53d6\u5f0f\u6458\u8981\u6210\u7e3e\u3002\u4e0a\u8ff0 \u4e4b\u8a2d\u5b9a\u4f9d\u7136\u7686\u9069\u7528\u5728 GALs \u7684\u62bd\u53d6\u5f0f\u6458\u8981\u5668\u4e0a\uff0c\u7136\u800c\u6211\u5011\u5728\u5be6\u9a57\u904e\u7a0b\u4e2d\u767c\u73fe(\u898b\u8868 \u8868\u4e00\u3001\u5404\u5f0f\u6a21\u578b\u65bc CNN/Daily Mail \u6578\u64da\u96c6\u4e4b\u6458\u8981\u7d50\u679c ROUGE-1 ROUGE-2 ROUGE-L Lead-3 40.34 17.70 36.57 Lead-3 (our implementation) 40.27 17.73 36.48 Baseline System [4] 40.17 18.11 36.41 Baseline System (our implementation) 39.69 17.91 36.01 REFRESH [14] 40.00 18.10 36.60 CRSum [2] 40.52 18.08 36.81 EXTRACT [15] 40.62 18.45 37.14 GALs 40.93 18.51 37.19 \u8868\u4e8c\u3001\u53c3\u8003\u6307\u6a19\u8207\u512a\u5316\u7d50\u679c ROUGE-1 ROUGE-2 ROUGE-L \u53c3\u8003\u6307\u6a19 ROUGE-1 40.93 18.51 37.19 ROUGE-2 40.77 18.52 37.07 ROUGE-L 40.48 18.39 36.80 \u5f0f\u6458\u8981\u6cd5\uff0c\u662f\u4e00\u7a2e\u5c0d\u62bd\u53d6\u5f0f\u6458\u8981\u5668\u505a\u6458\u8981\u5c64\u6b21\u512a\u5316\u7684\u65b9\u6cd5\u3002\u76f8\u8f03\u65bc\u66fe\u662f\u4e16\u754c\u6700\u4f73\u7d50\u679c \u4e4b\u7814\u7a76\u6240\u7528\u7684\u57fa\u6e96\u7cfb\u7d71\uff0c\u6211\u5011\u6240\u63d0\u51fa\u4e4b GALs \u80fd\u5728 CNN/Daily Mail \u6578\u64da\u96c6\u7684\u62bd\u53d6\u5f0f\u8207 \u91cd\u5beb\u5f0f\u6458\u8981\u7d50\u679c\u4e0a\uff0c\u6240\u6709 ROUGE \u6210\u7e3e\u7686\u7372\u5f97\u660e\u986f\u7684\u63d0\u6607\u3002\u6700\u5f8c\uff0c\u6211\u5011\u5c0d GALs \u6240\u4f7f\u7528 \u7684\u53c3\u6578\u8207\u7279\u5fb5\uff0c\u9032\u884c\u7d30\u90e8\u7684\u8abf\u67e5\u8207\u5206\u6790\uff0c\u85c9\u4ee5\u89c0\u5bdf\u767c\u60f3\u65bc\u8cc7\u6599\u6aa2\u7d22\u4e4b\u7814\u7a76\u800c\u8a2d\u8a08\u7684 GALs\uff0c\u5728\u512a\u5316\u62bd\u53d6\u5f0f\u6458\u8981\u6642\u6240\u5c55\u73fe\u7684\u7279\u6027\u8207\u6548\u80fd\u3002 \u81f4\u8b1d This work is supported by the Ministry of Science and Technology (MOST) in Taiwan under grant MOST 108-2636-E-011-005 (Young Scholar Fellowship Program), and by the Project J367B83100 (ITRI) under the sponsorship of the Ministry of Economic Affairs, Taiwan. \u53c3\u8003\u6587\u737b [1] T.-Y. Liu, \"Learning to rank for information retrieval,\" Foundations and Trends in Information Retrieval, vol. 3, no. 3, pp. 225-331, 2009. \u512a\u5316\u53e5\u6578 ROUGE-1 ROUGE-2 ROUGE-L 3 39.97 17.98 36.34 4 40.34 18.26 36.69 5 40.55 18.39 36.89 6 40.48 18.39 36.80 [2] P. \u8868\u4e09\u3001\u512a\u5316\u53e5\u6578\u8207\u512a\u5316\u7d50\u679c 7 40.40 18.33 36.72</td></tr><tr><td>\u4e09)\uff0c\u8b93\u62bd\u53d6\u5f0f\u6458\u8981\u5668\u56fa\u5b9a\u62bd\u53d6\u524d\u4e94\u500b\u53e5\u5b50\u4ee5\u505a\u6458\u8981\u5c64\u6b21\u7684\u5168\u5c40\u512a\u5316\uff0c\u80fd\u5927\u5e45\u63d0\u6607\u62bd</td></tr><tr><td>\u5fb5\uff0c\u6700\u5f8c\u6311\u9078\u51fa\u4ee5\u4e0b\u516b\u7a2e\u7279\u5fb5\uff1a \u512a\u826f\u7684\u6458\u8981\u61c9\u907f\u514d\u5197\u9918\u7684\u8a9e\u53e5\uff0c\u610f\u5373\u4efb\u5169\u500b\u53e5\u5b50\u9593\u90fd\u4e0d\u8a72\u64c1\u6709\u904e\u9ad8\u7684\u76f8\u4f3c\u6027\u3002\u57fa\u65bc\u53e5 \u53d6\u5f0f\u6458\u8981\u5668\u7684\u8868\u73fe\u3002\u6211\u5011\u8a8d\u70ba\u539f\u56e0\u5728\u65bc\u8a9e\u6599\u4e2d\u6709\u4e0d\u5c11\u6587\u672c\uff0c\u5176\u53c3\u8003\u6458\u8981\u7531\u8d85\u904e\u56db\u500b\u53e5</td></tr><tr><td>1\u3001\u53e5\u5b50\u9577\u5ea6 \u7df4\u6642\u7684\u4e0d\u7a69\u5b9a\u6027\u8207\u983b\u7e41\u7684\u68af\u5ea6\u7206\u70b8\uff0c\u6211\u5011\u63a1\u7528 Adam \u512a\u5316\u5668\u8207 0.00001 \u7684\u5b78\u7fd2\u7387\uff0c\u4e26\u88c1 \u5d4c\u5165\u7a7a\u9593\u7684\u7279\u6027\uff0c\u6211\u5011\u8a08\u7b97\u6458\u8981\u4e2d\u4efb\u5169\u500b\u53e5\u5b50\u9593\u53e5\u5d4c\u5165\u5411\u91cf\u7684\u9918\u5f26\u76f8\u4f3c\u5ea6\uff0c\u4ee5\u8a55\u4f30\u4efb \u5b50\u6240\u7d44\u6210\uff0c\u7576\u6211\u5011\u5c07\u62bd\u53d6\u4e4b\u7b2c\u56db\u8207\u7b2c\u4e94\u53e5\u8a71\u4e5f\u7d0d\u5165\u512a\u5316\u7684\u5c0d\u8c61\u4e2d\uff0c\u80fd\u4f7f\u62bd\u53d6\u5f0f\u6458\u8981\u5668</td></tr><tr><td>\u5b78\u7fd2\u7387\u8870\u6e1b\u7684\u6642\u6a5f\u3002\u5f9e\u8868\u4e00\u4e4b\u5be6\u9a57\u7d50\u679c\u53ef\u4ee5\u770b\u5230\uff0c\u7d93\u512a\u5316\u5f8c\u7684\u6a21\u578b\u5728\u62bd\u53d6\u5f0f\u6458\u8981\u7684\u8868 \u9577\u5ea6\u8f03\u9577\u7684\u53e5\u5b50\u542b\u6709\u66f4\u591a\u7684\u8a5e\u5f59\uff0c\u53ef\u80fd\u6db5\u84cb\u7684\u8cc7\u8a0a\u91cf\u4e5f\u8f03\u591a\uff0c\u56e0\u6b64\u8f03\u9577\u7684\u53e5\u5b50\u901a\u5e38\u8f03 \u526a\u68af\u5ea6\u81f3 2.0\u3002\u6211\u5011\u5728\u9a57\u8b49\u96c6\u4e0a\u6839\u64da ROUGE-1\uff0c\u81ea\u52d5\u6c7a\u5b9a\u65e9\u505c\u6cd5(Early Stopping)\u8207 \u5169\u500b\u53e5\u5b50\u9593\u7684\u76f8\u4f3c\u6027\uff0c\u6700\u5f8c\u4ee5\u5e73\u5747\u503c\u8207\u6a19\u6e96\u5dee\u4f5c\u70ba\u7279\u5fb5\u4f7f\u7528\u3002 \u5c0d\u9019\u4e9b\u6587\u672c\u751f\u6210\u66f4\u512a\u8cea\u4e4b\u6458\u8981\uff0c\u540c\u6642\u4ea6\u4e0d\u6703\u904e\u5ea6\u640d\u50b7\u62bd\u53d6\u4e4b\u524d\u4e09\u53e5\u8a71\u7684\u54c1\u8cea\u3002</td></tr><tr><td>(\u56db)\u6a21\u578b\u7d30\u7bc0 \u70ba\u91cd\u8981\uff0c\u7136\u800c\u62bd\u53d6\u904e\u9577\u7684\u53e5\u5b50\u4e5f\u53ef\u80fd\u9020\u6210\u6458\u8981\u7684\u5197\u9918\u6027\u63d0\u6607\u3002 \u73fe\u4e0a\uff0c\u6240\u6709 ROUGE \u6210\u7e3e\u7686\u6709\u660e\u986f\u7684\u63d0\u6607\u3002 \u4e94\u3001\u7d50\u8ad6</td></tr><tr><td>2\u3001\u7d55\u5c0d\u4f4d\u7f6e \u5728\u8a13\u7df4\u6642\uff0cGALs \u5c07\u65bc\u751f\u6210\u5668\u8207\u5224\u5225\u5668\u4ea4\u66ff\u8a13\u7df4\uff0c\u5224\u5225\u5668\u9700\u8981\u4e00\u7d44\u53c3\u8003\u6458\u8981(\u6b63\u6a23\u672c) \u8207\u751f\u6210\u5668\u6240\u751f\u6210\u4e4b\u6458\u8981(\u8ca0\u6a23\u672c)\u4f5c\u70ba\u8f38\u5165\uff0c\u4ee5\u4f5c\u4e8c\u5143\u5206\u985e\u7684\u8a13\u7df4\uff0c\u5176\u4e2d\u6b63\u8ca0\u6a23\u672c\u7684 (\u4e8c)\u65e9\u505c\u6cd5\u8207\u5b78\u7fd2\u7387\u8870\u6e1b \u6211\u5011\u5c07\u62bd\u53d6\u5f0f\u6458\u8981\u554f\u984c\u8996\u70ba\u6574\u5217\u5f0f\u6587\u672c\u6392\u5e8f\u554f\u984c\uff0c\u63d0\u51fa GALs\uff1a\u57fa\u65bc\u5c0d\u6297\u5f0f\u5b78\u7fd2\u4e4b\u6574\u5217</td></tr><tr><td>\u6587\u672c\u4e2d\u7684\u524d\u4e09\u81f3\u524d\u516d\u53e5\u8a71\u901a\u5e38\u6709\u9ad8\u5ea6\u7684\u91cd\u8981\u6027\uff0c\u56e0\u6b64\u53e5\u5b50\u5728\u6587\u672c\u4e2d\u7684\u7d55\u5c0d\u4f4d\u7f6e\u4e5f\u662f\u5224 \u5225\u5668\u7684\u8f38\u5165\u4e4b\u4e00\u3002 \u6a19\u6e96\u7b54\u6848\u5206\u5225\u70ba 1.0 \u8207 0.0\u3002\u56e0\u70ba\u8f38\u5165\u4e4b\u6a23\u672c\u5fc5\u70ba\u4e00\u6b63\u4e00\u8ca0\uff0c\u6211\u5011\u6368\u68c4\u50b3\u7d71\u7684\u4e8c\u5143\u4ea4\u53c9 \u5728 GALs \u512a\u5316\u6a21\u578b\u7684\u904e\u7a0b\u4e2d\uff0c\u6211\u5011\u89c0\u5bdf\u6a21\u578b\u5728\u9a57\u8b49\u96c6\u4e0a ROUGE-1\u3001ROUGE-2 \u8207</td></tr></table>", | |
| "text": "Ren, Z. Chen, Z. Ren, F. Wei, J. Ma, M. de Rijke, \"Leveraging contextual sentence relations for extractive summarization using a neural attention model,\" Proceedings of the 40th International ACM SIGIR Conference on Research and Development in Information Retrieval, pp. 95-104, 2017. [3] O. Vinyals, M. Fortunato, N. Jaitly, \"Pointer networks,\" Proceedings of the 28th International Conference on Neural Information Processing Systems, vol. 2, pp. 2692-2700, 2015.[4] Y.-C. Chen, M. Bansal, \"Fast abstractive summarization with reinforce-selected sentence rewriting,\" Proceedings of the 56th Annual Meeting of the Association for Computational Linguistics, vol. 1, pp. 675-686, 2018.", | |
| "html": null | |
| } | |
| } | |
| } | |
| } |