| 语义角色标注中特征优化组合研究 |
| |
| 引用本文： | 丁金涛,王红玲,周国栋,朱巧明,钱培德.语义角色标注中特征优化组合研究[J].计算机应用与软件,2009,26(5). |
| |
| 作者姓名： | 丁金涛 王红玲 周国栋 朱巧明 钱培德 |
| |
| 作者单位： | 江苏省计算机信息处理技术重点实验室,江苏,苏州,215006;苏州大学计算机科学与技术学院,江苏,苏州,215006 |
| |
| 基金项目： | 国家高技术研究发展计划(863计划)，国家自然科学基金，高等学校博士学科点专项科研基金 |
| |
| 摘 要： | 特征决定着语义角色标注的性能,但并非特征越多性能越高.因此,如何优化组合这些特征就显得非常重要.选取了当前语义角色标注系统中常用的56个特征,按其贡献进行了优化组合实验.在正确的句法分析上取得了91.22%的语义角色分类精确率,在自动句法分析上语义角色标注F1值达到了78.07%.实验表明提高性能并不需要很多特征,关键在于特征的优化组合;同时在句法分析中起重要作用的谓词和中心词特征在语义角色标中也发挥了重要作用.
|
| 关 键 词： | 语义角色标注 特征工程 最大熵分类器 |
ON OPTIMIZED COMBINATION OF FEATURES IN SEMANTIC ROLE LABELLING |
| |
| Ding Jintao,Wang Hongling,Zhou Guodong,Zhu Qiaoming,Qian Peide.ON OPTIMIZED COMBINATION OF FEATURES IN SEMANTIC ROLE LABELLING[J].Computer Applications and Software,2009,26(5). |
| |
| Authors: | Ding Jintao Wang Hongling Zhou Guodong Zhu Qiaoming Qian Peide |
| |
| Affiliation: | School of Computer Science and Technology;Soochow University;Suzhou 215006;Jiangsu;China;Jiangsu Provincial Key Laboratory of Computer Information Processing Technology;China |
| |
| Abstract: | Feature plays a critical role in performance of Semantic Role Labelling(SRL).One fact is that more features don't necessarily mean higher performance.Therefore,the proper optimized combination of the features becomes extremely important in SRL.In this paper it systematically explores fifty-six features which are frequently used in SRL and makes the experiment of optimized combination according to their evaluated contributions to SRL.Experiments on the CoNLL'2005 shared task in SRL show that our system achie... |
| |
| Keywords: | Semantic role labelling Feature engineering Maximum entropy classifier |
| 本文献已被 CNKI 维普 万方数据 等数据库收录！ |
