基于词语相似度的语义选择限制知识获取

Selectional Preference Acquisition Based on Word Similarity

考察基于词语相似度的语义选择限制知识获取方法.首先获取种子论元,再利用词语相似度进行论元扩展.比较了基于词向量的词语相似度计算方法与基于词典的词语相似度计算方法在选择限制知识获取任务中的表现.实验表明,前者效果更好,且二者有一定互补性,可以结合使用.与语义选择限制知识获取的其他方法相比,基于词语相似度的方法,种子选取灵活,不要求对语料进行深层句法语义分析,在伪消歧实验中也取得了较高的正确率,具有一定的优势.

Word similarity based on selectional preference acquisition method was investigated.Acquiring argument seeds was implemented firstly, and then they were expanded based on word similarity.Word similarity computing methods, word embedding based method, and dictionary based method were compared.It was shown that the former performed much better than the latter and they were complementary and could be combined.Word similarity based on selectional preference acquisition method had strength over other methods for its flexible seed choosing, no need of deep syntactic and semantic text processing, and high accuracy in pseudo-disambiguation experiments.