基于角度—相似度转换模型的义原相似度计算

The calculation of sememe similarity based on angle-similarity transformation model

义原的相似度是基于《知网》计算词语、句子相似度的基础.最早的义原相似度计算方法是刘群等提出的根据义原距离转换为义原相似度的方法,之后许多研究者在计算公式中引入了其他影响相似度计算的因素.这些因素本质上都可以由父节点深度、父节点到第一个义原的深度和父节点到第二个义原的深度等三个基本因素组成.将这三个因素作为基本变量,分析变量对义原相似度的影响方式并构建角度—相似度转换模型,实现了根据义原角度转换为义原相似度的方法.最后,通过实验分析比较新方法与同类方法在区分度、正区分度和距离误差上的优势,验证了其合理性.

Sememe similarity is the basis for computing similarity of words and sentences in the“Hownet”. The earliest sememe similarity calculation method was proposed by Liu Qu. In thismethod, the Iememe diItance waItranIformed into Iememe Iimilarity. After that, manyresearchers have introduced other factors that influence similarity calculations in the calculationformula. These factors essentially consists of three preliminary parts： depth of common fatherpoint, distance between the father point and the first sememe, and distance between the father point and the second sememe. Taking these three factors as basic variables, the effect of thesefactors on sememe similarity was analyzed and an angle-similarity transformation model was constructed to transform sememe angle into sememe similarity. Finally, the experiments analyzed the advantages on distinction, positive distinction and distance error between new methods andsimilar methods. and verified its rationality.