[1]田生伟,胡伟,禹龙,等.结合注意力机制的Bi-LSTM维吾尔语事件时序关系识别[J].东南大学学报(自然科学版),2018,48(3):393-399.[doi:10.3969/j.issn.1001-0505.2018.03.003]
 Tian Shengwei,Hu Wei,Yu Long,et al.Temporal relation identification of Uyghur event based on Bi-LSTM with attention mechanism[J].Journal of Southeast University (Natural Science Edition),2018,48(3):393-399.[doi:10.3969/j.issn.1001-0505.2018.03.003]
点击复制

结合注意力机制的Bi-LSTM维吾尔语事件时序关系识别()
分享到:

《东南大学学报(自然科学版)》[ISSN:1001-0505/CN:32-1178/N]

卷:
48
期数:
2018年第3期
页码:
393-399
栏目:
计算机科学与工程
出版日期:
2018-05-20

文章信息/Info

Title:
Temporal relation identification of Uyghur event based on Bi-LSTM with attention mechanism
作者:
田生伟1胡伟1禹龙1吐尔根·依布拉音2赵建国3李圃3
1新疆大学软件学院, 乌鲁木齐 830008; 2新疆大学信息科学与工程学院, 乌鲁木齐 830046; 3新疆大学中国语言学院, 乌鲁木齐 830046
Author(s):
Tian Shengwei 1 Hu Wei1 Yu Long1 Turgun Ibrayim2 Zhao Jianguo3 Li Pu3
1College of Software, Xinjiang University, Urumqi 830008, China
2College of Information Science and Technology, Xinjiang University, Urumqi 830046, China
3College of Chinese Language, Xinjiang University, Urumqi 830046, China
关键词:
维吾尔语 时序关系 注意力机制 双向长短时记忆网络 词向量
Keywords:
Uyghur temporal relation attention mechanism bidirectional-long short-term memory network word embedding
分类号:
TP391
DOI:
10.3969/j.issn.1001-0505.2018.03.003
摘要:
针对维吾尔语事件时序关系识别问题,提出了一种结合注意力机制的双向长短时记忆模型.基于维吾尔语语言及事件时序关系的特点,抽取13项基于事件间内部结构信息的特征.将词向量作为双向长短时记忆模型的输入,挖掘给定事件句隐含的上下文语义信息.结合事件触发词建立注意力机制,获取该事件句的事件语义特征.将事件内部结构特征和语义特征相融合,作为softmax层的输入,进而完成事件时序关系的识别.实验结果表明,该方法在获取事件句隐含语义信息的同时也能获取对应的事件语义特征.融合事件内部结构特征后,识别准确率为89.42%,召回率为86.70%,衡量模型整体性能的F值为88.03%,从而证明了该方法在维吾尔语事件时序关系识别任务上的有效性.
Abstract:
As for the Uyghur event temporal relation identification problem, a model based on bidirectional-long short-term memory(Bi-LSTM)with attention mechanism is proposed. Based on the characteristics of Uyghur language and event temporal relation, 13 features of event internal structural information are extracted. The word embedding is introduced as the Bi-LSTM input to mine the context semantic information implied by a given event sentence. An attention mechanism is established with the event triggers to obtain the event semantic features of the given event sentence. The event internal structural features and the semantic features are combined to be the input of the softmax layer to complete the identification of event temporal relation. The experimental results show that the method can obtain the semantic information of the context and the implicit semantic features of the corresponding event sentence. After fusing the internal structural characteristics of the event, the identification precision rate is 89.42%; the recall rate is 86.70% and the F value for measuring the overall performance of the model is 88.03%,indicating the effectiveness of this method in the identification task of Uyghur event temporal relation.

参考文献/References:

[1] 孙辉. 事件时序关系识别的研究与实现[D]. 哈尔滨:哈尔滨工业大学计算机科学与技术学院,2010.
[2] Tourille J, Ferret O, Neveol A, et al. Neural architecture for temporal relation extraction: A Bi-LSTM approach for detecting narrative containers [C] //Proc of the 55th Annual Meeting of the Association for Computational Linguistics(Short Papers). Vancouver, Canada, 2017: 224-230. DOI:10.18653/v1/p17-2035.
[3] Styler W, Bethard S, Finan S, et al. Temporal annotation in the clinical domain [C] //The 52nd Annual Meeting of the Association for Computational Linguistics. Baltimore, Maryland, USA, 2014: 143-154.
[4] 郑新, 李培峰, 朱巧明. 基于全局优化的中文事件时序关系推理方法[J]. 中文信息学报, 2016, 30(5): 129-135.
  Zheng Xin,Li Peifeng,Zhu Qiaoming.Global inference for temporal relations between Chinese events[J].Journal of Chinese Information Processing,2016,30(5):129-135.(in Chinese)
[5] Tang D, Qin B,Feng X, et al. Effective LSTMs for target-dependent sentiment classification [C] //Proc of COLING 2016. Osaka, Japan, 2016: 3298-3307.
[6] Zhou X, Wan X, Xiao J. Attention-based LSTM network for cross-lingual sentiment classification [C] //Proceedings of the 2016 Conference on Empirical Methods in Natural Language Processing. Austin, Texas, USA, 2016: 247-256. DOI:10.18653/v1/d16-1024.
[7] 付剑锋. 面向事件的知识处理研究[D].上海:上海大学计算机学院,2010.
[8] 钟军, 禹龙, 田生伟,等. 基于双层模型的维吾尔语突发事件因果关系抽取[J]. 自动化学报, 2014, 40(4):771-779. DOI: 10.3724/SP.J.1004.2013.00771.
Zhong Jun,Yu Long,Tian Shengwei,et al.Causal relation extraction of Uyghur emergency events based on cascaded model[J].Acta Automatica Sinica,2014,40(4):771-779. DOI:10.3724/SP.J.1004.2013.00771. (in Chinese)
[9] Hochreiter S, Schmidhuber J. Long short-term memory[J]. Neural Computation, 1997, 9(8): 1735-1780.
[10] Gers F A, Schmidhuber J. Recurrent nets that time and count [C] //Proc of IEEE-INNS-ENNS Int Joint Conf on Neural Networks. Como, Italy, 2000: 189-194.
[11] Melamud O, Goldberger J, Dagan I. Context2vec: Learning generic context embedding with bidirectional LSTM[C]// Proceedings of the 20th SIGNLL Conference on Computational Natural Language Learning. Berlin, Germany, 2016:51-61. DOI:10.18653/v1/k16-1006.
[12] Rocktäschel T, Grefenstette E, Hermann K M, et al. Reasoning about entailment with neural attention [C] //Proc of Int Conf on Learning Representations. San Juan, Puerto Rico, 2016:1-9.
[13] 李敏, 禹龙, 田生伟,等. 基于深度学习的维吾尔语名词短语指代消解[J]. 自动化学报, 2017(11):1984-1992. DOI: 10.16383/j.aas.2017.c160330.
Li Min,Yu Long,Tian Shengwei,et al. Coreference resolution of Uyghur noun phrases based on deep learning[J].Acta Automatica Sinica,2017(11):1984-1992. DOI:10.16383/j.aas.2017.c160330. (in Chinese)
[14] 田生伟, 周兴发, 禹龙,等. 基于双向LSTM的维吾尔语事件因果关系抽取[J]. 电子与信息学报, 2018, 40(1):200-208. DOI: 10.11999/JEIT170402.
Tian Shengwei,Zhou Xingfa,Yu Long,et al.Causal relation extraction of Uyghur events based on bidirectional long short-term memory model[J].Journal of Electronics and Information Technology,2018,40(1):200-208. DOI:10.11999/JEIT170402. (in Chinese)

备注/Memo

备注/Memo:
收稿日期: 2017-11-26.
作者简介: 田生伟(1973—),男,博士,教授, tianshengwei@163.com.
基金项目: 国家自然科学基金资助项目(61262064,61331011,61563051,61662074)、新疆维吾尔自治区科技人才培养资助项目(QN2016YX005).
引用本文: 田生伟,胡伟,禹龙,等.结合注意力机制的Bi-LSTM维吾尔语事件时序关系识别[J].东南大学学报(自然科学版),2018,48(3):393-399. DOI:10.3969/j.issn.1001-0505.2018.03.003.
更新日期/Last Update: 2018-05-20