一个基于超图的词义归纳模型

A Hypergraph Model for Word Sense Induction

针对词义归纳如何学习多个上下文实例中的高阶语义关系的问题,提出一个基于超图的词义归纳模型。首先,采用基于词汇链的方法发现目标单词的上下文实例间的高阶语义关系;然后,用上下文实例表示结点,用词汇链发现超边来构建超图;最后,使用一个基于最大密度超图谱聚类算法发现词义。实验基于Semeval-2013 WSI任务,与普通图模型进行比较,其在词义检测与词义评级2个指标上分别提升了5.6%和6.4%。

In order to learn the higher-order semantic relatedness among multiple instance of target word,a hypergraph_model was proposed for word sense induction. First,a lexical chain based method was used for discovering the higher-order semantic relatedness. Then a hypergraph was constructed,in which nodes represent the instances of contexts where a target word occurs,and hyperedges were formed by lexical chains. Finally,a maximum density based hypergraph clustering method was used for finding word senses. Experiments based on Semeval-2013 WSI task showed that this model gives an improvement of 5. 6% and 6. 4% in sense detection and sense ranking respectively,compared to the traditional graph model.