A Multi-Mechanism Fused Paraphrase Generation Model with Joint Auto-Encoding Learning

doi:10.13209/j.0479-8023.2019.104

Acta Scientiarum Naturalium Universitatis Pekinensis ›› 2020, Vol. 56 ›› Issue (1): 53-60.DOI: 10.13209/j.0479-8023.2019.104

Previous Articles Next Articles

A Multi-Mechanism Fused Paraphrase Generation Model with Joint Auto-Encoding Learning

LIU Mingtong¹, ZHANG Yujie^1,†, ZHANG Shu², MENG Yao³, XU Jin’an¹, CHEN Yufeng¹

1. School of Computer and Information Technology, Beijing Jiaotong University, Beijing 100044 2. Fujitsu Research and Development Center, Beijing 100027 3. Lenovo Research, AI Lab, Beijing 100085

Received:2019-05-23 Revised:2019-09-20 Online:2020-01-20 Published:2020-01-20
Contact: ZHANG Yujie, E-mail: yjzhang(at)bjtu.edu.cn

联合自编码任务的多机制融合复述生成模型

刘明童¹, 张玉洁^1,†, 张姝², 孟遥³, 徐金安¹, 陈钰枫¹

1. 北京交通大学计算机与信息技术学院, 北京 100044 2. 富士通研究开发中心, 北京 100027 3. 联想研究院人工智能实验室, 北京 100085

通讯作者: 张玉洁, E-mail: yjzhang(at)bjtu.edu.cn
基金资助:
国家自然科学基金(61876198, 61976015, 61370130, 61473294)、中央高校基本科研业务费专项资金(2018YJS025)、北京市自然科学基金(4172047)和科学技术部国际科技合作计划(K11F100010)资助

Abstract

Abstract:

Neural network encoder-decoder framework has become the popular method for paraphrase generation, but there are still two problems. On the one hand, there are such issues as inaccurate entity words, unknown words and word repetition in the generated paraphrase sentences. To solve the first problem, we proposed a multimechanism fused paraphrase generation model to improve the decoder. The copy mechanism was used to copy words form input sentence for improving the generation of entity and unknown words. The coverage mechanism was used to model historical attention information to avoid word repetition. On the other hand, the limited-scale parallel paraphrase corpus limits the learning ability of the encoder. We proposed to jointly learn auto-encoding task, which shares one encoder with paraphrase generation task. The joint auto-encoding task enhances the learning ability of the encoder. Experimental results on Quora paraphrase dataset show that the multi-mechanism fused paraphrase generation model with joint auto-encoding task can effectively improve the performance of paraphrase generation.

Key words: paraphrase generation, auto-encoding, multi-task learning, multi-mechanism fusion, attention mechanism, copy mechanism, coverage mechanism

摘要：

基于神经网络编码–解码框架的复述生成模型存在两方面的问题: 1) 生成的复述句中存在实体词不准确、未登录词和词汇重复生成; 2) 复述平行语料的有限规模限制了编码器的语义学习能力。针对第一个问题, 本文提出在解码过程中融合注意力机制、复制机制和覆盖机制的多机制复述生成模型, 利用复制机制从原句复制词语来解决实体词和未登录词生成问题; 利用覆盖机制建模学习注意力机制历史决策信息来规避词汇重复生成。针对第二个问题, 基于多任务学习框架, 提出在复述生成任务中联合自编码任务, 两个任务共享一个编码器, 同时利用平行复述语料和原句子数据, 共同增强复述生成编码器的语义学习能力。在Quora复述数据集上的实验结果表明, 提出的联合自编码的多机制融合复述生成模型有效地解决了复述生成的问题, 并提高了复述句的生成质量。

关键词: 复述生成, 自编码, 多任务学习, 多机制融合, 注意力机制, 复制机制, 覆盖机制

LIU Mingtong, ZHANG Yujie, ZHANG Shu, MENG Yao, XU Jin’an, CHEN Yufeng. A Multi-Mechanism Fused Paraphrase Generation Model with Joint Auto-Encoding Learning[J]. Acta Scientiarum Naturalium Universitatis Pekinensis, 2020, 56(1): 53-60.

刘明童, 张玉洁, 张姝, 孟遥, 徐金安, 陈钰枫. 联合自编码任务的多机制融合复述生成模型[J]. 北京大学学报自然科学版, 2020, 56(1): 53-60.

Add to citation manager EndNote|Ris|BibTeX

URL: https://xbna.pku.edu.cn/EN/10.13209/j.0479-8023.2019.104

https://xbna.pku.edu.cn/EN/Y2020/V56/I1/53

[1]	LI Chao, HOU Xia, QIAO Xiuming. A Low-Resource Named Entity Recognition Method for Cultural Heritage Field Incorporating Knowledge Fusion [J]. Acta Scientiarum Naturalium Universitatis Pekinensis, 2024, 60(1): 13-22.
[2]	WANG Rui, LIANG Jianguo, HUA Rong. Radiology Report Generation Method Based on Multi-scale Feature Parsing [J]. Acta Scientiarum Naturalium Universitatis Pekinensis, 2024, 60(1): 71-78.
[3]	DING Na, LIU Peng, SHAO Huipeng, WANG Xuekui. Bi-Attention Text-Keyword Matching for Law Recommendation [J]. Acta Scientiarum Naturalium Universitatis Pekinensis, 2024, 60(1): 79-88.
[4]	TIAN Yuan, PANG Xiao, ZHAO Wenyi, CHANG Xiaoyin, CHENG Chuyun, ZOU Pei, CAO Xiaocheng. A Transformer-Based Model for Short-Term Landslide Displacement Prediction [J]. Acta Scientiarum Naturalium Universitatis Pekinensis, 2023, 59(2): 197-210.
[5]	GAO Xiaoqian, ZHOU Xiabing, ZHANG Min. A Multi-information Perception Based Method for Question Answering in Multi-party Conversation [J]. Acta Scientiarum Naturalium Universitatis Pekinensis, 2023, 59(1): 21-29.
[6]	CHEN Yuan, QIU Xinying. Multi-task Semantic Matching with Self-supervised Learning [J]. Acta Scientiarum Naturalium Universitatis Pekinensis, 2022, 58(1): 83-90.
[7]	SU Fenglong, SUN Chengzhe, JING Ning. A Context-Fusion Method for Entity Extraction Based on Residual Gated Convolution Neural Network [J]. Acta Scientiarum Naturalium Universitatis Pekinensis, 2022, 58(1): 69-76.
[8]	MA Chao, WAN Zhang, ZHANG Yujie, XU Jin’an, CHEN Yufeng. Multi-modality Paraphrase Generation Model Integrating Image Information [J]. Acta Scientiarum Naturalium Universitatis Pekinensis, 2022, 58(1): 45-53.
[9]	WANG Yinmiao, HAN Zhimin. Knowledge Bases Completion Based on Multi-hop Paths [J]. Acta Scientiarum Naturalium Universitatis Pekinensis, 2022, 58(1): 7-12.
[10]	WAN Zhang, ZHANG Yujie, LIU Mingtong, XU Jin’an, CHEN Yufeng. Object Space Relation Mechanism Fused Image Caption Method [J]. Acta Scientiarum Naturalium Universitatis Pekinensis, 2021, 57(1): 75-82.
[11]	LIN Zijie, LONG Yunfei, DU Jiachen, XU Ruifeng. A Multi-modal Sentiment Recognition Method Based on Multi-task Learning [J]. Acta Scientiarum Naturalium Universitatis Pekinensis, 2021, 57(1): 7-15.
[12]	WU Ruiying, KONG Fang. Event Coreference Resolution with Document Representation [J]. Acta Scientiarum Naturalium Universitatis Pekinensis, 2020, 56(1): 82-88.
[13]	JIA Yunlong, HAN Donghong, LIN Haiyuan, WANG Guoren, XIA Li. Consumption Intent Recognition Algorithms for Weibo Users [J]. Acta Scientiarum Naturalium Universitatis Pekinensis, 2020, 56(1): 68-74.
[14]	HUANG Yan, SUN Haili, XU Ke, YU Xiaoyang, WANG Tongyang, ZHANG Xinfang, LU Songfeng. Discourse-Level Text Generation Method Based on Topical Constraint [J]. Acta Scientiarum Naturalium Universitatis Pekinensis, 2020, 56(1): 9-15.
[15]	LI Lanjun, ZHOU Junsheng, GU Yanhui, Qü Weiguang. Similar Legal Case Retrieval Based on Improved Siamese Network [J]. Acta Scientiarum Naturalium Universitatis Pekinensis, 2019, 55(1): 84-90.

A Multi-Mechanism Fused Paraphrase Generation Model with Joint Auto-Encoding Learning

联合自编码任务的多机制融合复述生成模型

PDF

Knowledge

Abstract

Cite this article

share this article

References

Related Articles 15

Recommended Articles

Metrics