多特征结合的词语相似度计算模型

词语相似度计算在基于实例的机器翻译、信息检索、自动问答系统等有着广泛的应用。词语相似度的计算一般都是在基于《知网》的义原的基础上,通过计算概念之间的相似度来获取。文中在综合考虑义原距离、义原深度、义原宽度、义原密度和义原重合度的基础上,利用多特征结合的方法计算词语相似度。为了验证算法的合理性,利用Miller和Charles文献给出的基准词作为测试集合,将计算得到的词语相似度的值与专家值进行比较,计算其皮尔逊相关系数,计算结果达到了0.852。实验结果表明多特征结合的词语相似度计算和专家评定的词语相似度计算非常吻合。

Word Similarity Computation Model of Multi-features Combination

Semantic similarity computing has been widely used in machine translation based on example,information retrieval and automatic question answering systems. Word similarity computation is generally based on the original in ＂ HowNet＂,through calculating the degree of similarity between concepts to obtain. In this paper,in consideration of the original distance,depth,width,density and contact ratio,use the method with multi- features to compute word similarity. In order to verify the rationality of the algorithm,using the benchmark of words given by M iller and Charles literature as a test set,make a comparison between the word similarity computation values and expert value,calculating the Pearson correlation coefficient,the calculation results is 0. 852. Experimental result showthat the word similarity computation of multi- features combination is identical with expert estimation.