基于测度优化Laplacian SVM的中文指代消解方法

doi:10.3969/j.issn.0372-2112.2016.12.035

电子学报 ›› 2016, Vol. 44 ›› Issue (12): 3064-3072.DOI: 10.3969/j.issn.0372-2112.2016.12.035

基于测度优化Laplacian SVM的中文指代消解方法

周炫余^1,2, 刘娟^1,2, 邵鹏^1,2, 卢笑³, 罗飞^1,2

1. 武汉大学软件国家重点实验室, 湖北武汉 430072;
2. 武汉大学计算机学院, 湖北武汉 430072;
3. 湖南大学电气与信息工程学院, 湖南长沙 410082

收稿日期:2015-11-05 修回日期:2016-06-30 出版日期:2016-12-25
- 通讯作者:
- 刘娟
- 作者简介:
- 周炫余,男,1987年10月出生,湖南邵阳人.现为武汉大学计算机学院博士研究生,主要从事指代消解、中文自然语言处理、机器学习等有关研究.E-mail:zhouxuanyu@whu.edu.cn
- 基金资助:
- 国家自然科学基金 (No.61272274）; 国家自然科学基金青年项目 (No.61402340）; 湖北省自然科学基金 (No.2014CFB194）

Chinese Anaphora Resolution Based on Metric-optimized Laplacian SVM

ZHOU Xuan-yu^1,2, LIU Juan^1,2, SHAO Peng^1,2, LU Xiao³, LUO Fei^1,2

1. State Key Laboratory of Software Engineering, Wuhan University, Wuhan, Hubei 430072, China;
2. Computer School, Wuhan University, Wuhan, Hubei 430072, China;
3. College of Electrical and Information Engineering, Hunan University, Changsha, Hunan 410082, China

Received:2015-11-05 Revised:2016-06-30 Online:2016-12-25 Published:2016-12-25
- Supported by:
- National Natural Science Foundation of China (No.61272274); Youth Fund of National Natural Science Foundation of China (No.61402340); Natural Science Foundation of Hubei Province, China (No.2014CFB194)

摘要/Abstract

摘要：

相比于传统的基于半监督学习的指代消解方法，Laplacian SVM（Support Vector Machine）能有效的挖掘已标注样本和未标注样本的相似性和关联性，更好的推导模型的分类边界.而传统Laplacian SVM采用欧式距离度量样本之间的距离，使得异类样本之间的相似性可能过大，不利于样本的准确分类.对此，提出一种基于数据驱动学习最优测度Laplacian SVM算法以解决中文指代消解语料不足的问题.该方法通过优化样本对之间的相似性约束条件和引入Fisher判别项，增大同类样本间的相似性，并突出强判别能力的特征.此外，提出核嵌入的测度优化方法将以上线性测度优化推广到非线性空间，有利于Laplacian SVM利用核函数实现非线性分类.在ACE2005中文语料库上的测评结果表明，所提出测度优化的Laplacian SVM（包括线性和核嵌入两种形式）的方法只需少量标注样本就可以获得与经典的有监督学习模型相当甚至更好的消解性能，同时也优于其他传统的半监督学习方法.

关键词: 测度优化, Laplacian SVM, 中文指代消解, 半监督学习, 自然语言处理

Abstract:

Compared to the traditional semi-supervised based anaphora resolution methods,Laplacian SVM(Support Vector Machine) can efficiently explore the similarity and correlations between labeled and unlabeled samples for deriving more accurate classification model.However,traditional Laplacian SVM simply uses Euclidean distance to calculate the distance between two samples,which may result that two samples from different classes may have false high similarity.To address the problem of insufficient Chinese annotated corpus,a data-driven based method is proposed to learn the optimal distance metric.The proposed method takes similarity constraints between sample-pairs into consideration and introduces the Fisher discrimination criterion,so that the similarities of in-class samples are higher than those of between-class samples,and the discriminant features are highlighted in the new metric space.Furthermore,the proposed metric-optimized method is generalized from linear to nonlinear space by the use of kernel,so that it can be used for non-linear classification.Compared with the classical supervised method and other four traditional semi-supervised methods on the ACE2005 Chinese corpus,the proposed method,both the linear form and kernel form,achieves the comparatively better or best performance,with fewer labeled samples.

Key words: metric-optimized, Laplacian SVM, Chinese anaphora resolution, semi-supervised learning, natural language processing

中图分类号:

TP391

周炫余, 刘娟, 邵鹏, 等. 基于测度优化Laplacian SVM的中文指代消解方法[J]. 电子学报, 2016, 44(12): 3064-3072.

ZHOU Xuan-yu, LIU Juan, SHAO Peng, et al. Chinese Anaphora Resolution Based on Metric-optimized Laplacian SVM[J]. Acta Electronica Sinica, 2016, 44(12): 3064-3072.

参考文献

[1] HARDMEIER C,FEDERICO M.Modelling pronominal anaphora in statistical machine translation[A].Proceedings of the International Workshop on Spoken Language Translation[C].Paris:Konferensbidrag,2010.283-289.
[2] 张志昌,张宇,刘挺,等.开放域问答技术研究进展[J].电子学报,2009,37(5):1058-1069. ZHANG Zhi-chang,ZHANG yu,LIU Ting,et al.Advances in open-domain question answering[J].Acta Electronica Sinica,2009,37(5):1058-1069.(in Chinese)
[3] 杨晓兰,钟义信.基于文本理解的自动文摘系统研究与实现[J].电子学报,1998,26(7):155-158. YANG Xiao-lan,ZHONG Yi-xin.Study and realization for text interpretation and automatic abstracting[J].Acta Electronica Sinica,1998,26(7):155-158.(in Chinese)
[4] 李维刚,刘挺,李生.基于网络挖掘的实体关系元组自动获取[J].电子学报,2007,35(11):2111-2116. LI Wei-gang,LIU Ting,LI Sheng.Automated entity relation tuple extraction using web mining[J].Acta Electronica Sinica,2007,35(11):2111-2116.(in Chinese)
[5] SOON W M,NG H T,LIm D C Y.A machine learning approach to coreference resolution of noun phrases[J].Computational Linguistics,2001,27(4):521-544.
[6] KEHLER A,APPELT D,TAYLOR L,et al.Competitive self-trained pronoun interpretation[A].Proceedings of HLT-NAACL[C].Stroudsburg,USA:ACL,2004.33-36.
[7] MVLLER C,RAPP S,STRUBE M.Applying co-training to reference resolution[A].Proceedings of the 40th Annual Meeting on ACL[C].Stroudsburg,USA:ACL,2002.352-359.
[8] RAGHAVAN P,FOSLER-LUSSIER E,LAI A M.Exploring semi-supervised coreference resolution of medical concepts using semantic and temporal features[A].Proceedings of the 2012 Conference of the NAACL[C].Stroudsburg,USA:ACL,2012.731-741.
[9] NG V,CARDIE C.Weakly supervised natural language learning without redundant views[A].Proceedings of the Conference of the NAACL[C].Stroudsburg,USA:ACL,2003.94-101.
[10] CHARNIAK E,ELSNER M.EM works for pronoun anaphora resolution[A].Proceedings of the 12th Conference of the EACL[C].Stroudsburg,USA:ACL,2009.148-156.
[11] BELKIN M,NIYOGI P,SINDHWANI V.Manifold regularization:A geometric framework for learning from labeled and unlabeled examples[J].The Journal of Machine Learning Research,2006,7(11):2399-2434.
[12] KULIS B.Metric learning:A survey[J].Foundations and Trends in Machine Learning,2012,5(4):287-364.
[13] STOYANOV V,GILBERT N,CARDIE C,et al.Conundrums in noun phrase coreference resolution:Making sense of the state-of-the-art[A].Proceedings of the Joint Conference of the 47th Annual Meeting of the ACL and the 4th International Joint Conference on Natural Language Processing of the AFNLP[C].Stroudsburg,USA:ACL,2009.656-664.
[14] 周炫余,刘娟,邵鹏,等.基于层次过滤模型的中文指代消解研究[J].吉林大学学报(工学版),2016,46(4):1209-1215. ZHOU Xuan-yu,LIU Juan,SHAO Peng,et al.Research of Chinese anaphora resolution based on multi-pass sieve model[J].Journal of Jilin University (Engineering and Technology Edition),2016,46(4):1209-1215.(in Chinese)
[15] CHIANG L H,RUSSELL E L,BRAATZ R D.Fisher Discriminant Analysis//Fault Detection and Diagnosis in Industrial Systems[M].London:Springer,2001.57-70.
[16] 孔芳,周国栋.基于树核函数的中英文代词消解[J].软件学报,2012,23(5):1085-1099. KONG Fang,ZHOU Guo-dong.Pronoun resolution in English and Chinese languages based on tree kernel[J].Journal of Software,2012,23(5):1085-1099.(in Chinese)
[17] JAIN P,KULIS B,DHILLON I S.Inductive regularized learning ofkernel functions[A].Proceedings of the NIPS[C].Massachusetts,USA:MIT Press,2010.946-954.
[18] MIKA S,RÄTSCH G,WESTON J,et al.Fisher discriminant analysis with kernels[A].Proceedings of the 1999 IEEE Signal Processing Society Workshop[C].Portland,USA:IEEE,1999.41-48.
[19] MARC V,JOHN B,JOHN A,et al.A model theoretic coreference scoring scheme[A].Proceedings of the 6th Message Understanding Conference[C].Stroudsburg,USA:ACL,1995.45-52.
[20] AMIT B,BRECK B.Algorithms for scoring coreference chains[A].Proceedings of LREC[C].Stroudsburg,USA:ACL,1998.563-566.
[21] LUO X Q.On coreference resolution performance metrics[A].Proceedings of HLT-EMNLP[C].Stroudsburg,USA:ACL,2005.25-32.
[22] NONG Y.The Handbook of Data Mining[M].Cleveland:CRC Press,2001.247-277.
[23] MARTA R,EDUARD H.BLANC:Implementing the Rand Index for coreference evaluation[J].Natural Language Engineering,2011,17(4):485-510.

基于测度优化Laplacian SVM的中文指代消解方法

Chinese Anaphora Resolution Based on Metric-optimized Laplacian SVM

PDF

可视化

摘要/Abstract

引用本文

使用本文

参考文献

相关文章 15

编辑推荐

Metrics

[1]	张志昌, 于沛霖, 庞雅丽, 朱林, 曾扬扬. SMGN：用于对话状态跟踪的状态记忆图网络[J]. 电子学报, 2022, 50(8): 1851-1858.
[2]	江兵兵, 何文达, 吴兴宇, 项俊浩, 洪立斌, 盛伟国. 基于自适应图学习的半监督特征选择[J]. 电子学报, 2022, 50(7): 1643-1652.
[3]	杜晋华, 尹浩, 冯嵩. 中文电子病历命名实体识别的研究与进展[J]. 电子学报, 2022, 50(12): 3030-3053.
[4]	黄名选. 关联模式挖掘与词向量学习融合的伪相关反馈查询扩展[J]. 电子学报, 2021, 49(7): 1305-1313.
[5]	张昱, 刘开峰, 张全新, 王艳歌, 高凯龙. 基于组合-卷积神经网络的中文新闻文本分类[J]. 电子学报, 2021, 49(6): 1059-1067.
[6]	杨启萌, 禹龙, 田生伟, 艾山·吾买尔. 基于深度强化学习的维吾尔语人称代词指代消解[J]. 电子学报, 2020, 48(6): 1077-1083.
[7]	黄名选, 蒋曹清. 基于项权值排序挖掘的跨语言查询扩展[J]. 电子学报, 2020, 48(3): 568-576.
[8]	吴玉佳, 李晶, 宋成芳, 常军. 基于高效用神经网络的文本分类方法[J]. 电子学报, 2020, 48(2): 279-284.
[9]	张志昌, 曾扬扬, 庞雅丽. 融合语义角色和自注意力机制的中文文本蕴含识别[J]. 电子学报, 2020, 48(11): 2162-2169.
[10]	马慧芳, 刘文, 李志欣, 蔺想红. 融合耦合距离区分度和强类别特征的短文本相似度计算方法[J]. 电子学报, 2019, 47(6): 1331-1336.
[11]	余游, 冯林, 王格格, 徐其凤. 一种基于伪标签的半监督少样本学习模型[J]. 电子学报, 2019, 47(11): 2284-2291.
[12]	黄健航, 雷迎科. 基于半监督矩形网络的通信电台个体识别[J]. 电子学报, 2019, 47(1): 1-8.
[13]	田生伟, 秦越, 禹龙, 吐尔根·依布拉音, 冯冠军. 基于Bi-LSTM的维吾尔语人称代词指代消解[J]. 电子学报, 2018, 46(7): 1691-1699.
[14]	马慧芳, 刘芳, 夏琴, 郝占军. 基于加权超图随机游走的文献关键词提取算法[J]. 电子学报, 2018, 46(6): 1410-1414.
[15]	黄名选, 蒋曹清. 基于完全加权正负关联模式挖掘的越-英跨语言查询译后扩展[J]. 电子学报, 2018, 46(12): 3029-3036.