【摘要】 提出一种基于条件随机场的领域术语上下位关系获取方法。首先,结合百科名片中结构化、制式化的语言表达形式,通过统计分析,提炼出适用于通用模型的特征词词典。然后,在词和词性特征的基础上,结合特征词词典内容和标点符号信息,利用CRF机器学习技术对术语间上下位关系的内在规律进行学习,得到其表达方式和存在环境的概率模型。最后,通过实验对模型的准确性进行验证,并提出了改进。实验结果表明:该方法抽取上下位关系的准确率达到73.50%。更多还原

【Abstract】 A method based on conditional random fields for automatic domain-specific terminological hyponymy extraction was proposed.First,taking the structured and regularized content expression forms of Baike card into consideration,a feature word dictionary that is suitable for general-purpose models after statistical analysis was summarized.Second,on the basis of the word and part of speech tagging(POS) features,combined with the feature word dictionary and punctuation,the inherent laws of domain-specific terminological hyponymy were learnt by CRF machine learning techniques,and a probabilistic model about the expression and the existing environment was obtained.At last,the accuracy of the model by means of a series of contrast experiments was verified and some improving schemes were put forward.The experimental results show that the accuracy rate of hyponymy extraction reaches 73.50% using the proposed method.更多还原