无指导学习语义优选  

UNSUPERVISED LEARNING OF SEMANTIC SELECTIONAL PREFERENCES

作　　者：李东明[1] 张丽娟[2] 赵伟[1] 石晶[2] 

机构地区：[1]吉林农业大学信息技术学院,吉林长春130118 [2]长春工业大学计算机科学与工程学院,吉林长春130012

出　　处：《计算机应用与软件》2012年第1期155-158,216,共5页Computer Applications and Software

基　　金：吉林省科技发展计划项目青年基金(20100155);吉林省科研发展计划科技支撑重点项目(20100214)

摘　　要：给出基于LSC模型的EM方法进行汉语语义优选的学习。具体步骤是首先随机为参数模型赋予初值;然后迭代运行EM算法,直到收敛;最后计算动词和名词的语义关联度,以此衡量其搭配的可能性。大量实验结果表明LSC模型能够较好地体现动、名词的搭配模式,且算法迭代收敛速度快。该方法无需语法标注的语料库,适合应用于汉语。An Expectation-Maximisation（EM） algorithm based on latent semantic clustering（LSC） model is introduced for learning Chinese semantic selectional preferences.The specific procedure is as follows： First,the model parameters are designated their initial values randomly;secondly,EM algorithm is executed iteratively until convergence achieved;finally,the semantic association between verbs and nouns is calculated to measure their collocation possibility.Lots of experiment results show that LSC model is able to provide proper collocation patterns of verbs and nouns and the iterative convergence speed of the algorithm is fast as well.The method is suitable for Chinese as it does not need syntax-annotated corpora.

关 键 词：语义优选 潜在语义聚类 无指导学习 

分 类 号：TP18[自动化与计算机技术—控制理论与控制工程]