基于贝叶斯估计的概念语义相似度算法  

A Concept Semantic Similarity Algorithm Based on Bayesian Estimation

作　　者：吴奎[1] 周献中[2] 王建宇[1] 赵佳宝[2] 

机构地区：[1]南京理工大学自动化学院,江苏南京210094 [2]南京大学工程管理学院,江苏南京210093

出　　处：《中文信息学报》2010年第2期52-57,67,共7页Journal of Chinese Information Processing

基　　金：国家自然科学基金资助项目(90718036)

摘　　要：传统的基于语义距离的概念语义相似度算法不能兼顾客观统计数据,基于信息量的相似度算法又难以获得权威统计样本,针对这些不足,该文提出一种基于贝叶斯估计的概念语义相似度算法。该算法首先假定概念出现概率是符合Beta分布的随机变量,然后基于语义距离的相似度算法计算先验参数,并根据统计样本计算该先验分布下基于最小风险的贝叶斯估计后验参数。随后利用基于信息量的语义相似度算法,便可获得主观经验与客观事实相结合的概念语义相似度。结合WordNet的实验分析表明,该算法与人为主观经验之间具有最大的相关系数。Traditional algorithms for semantic similarity computation fall into two categories： distance-based and in- formation-based methods. The former ignores the objective statistics, while the latter suffers from insufficient do- main data. In this paper, a new method for similarity computation based on Bayesian Estimation is proposed. First, the concept emergence probability is assumed to be a random variable with a priori Beta distribution. Second, its pri ori parameters are designated by the distance-based similarity algorithm, calculated by Bayesian Estimation. There- by, the semantic similarity integrating the subjective experience with the objective statistic is acquired based on in- formation-based method. Finally, the proposed method is implemented and proved by a slightly higher correlation with human judgments against WordNet.

关 键 词：计算机应用 中文信息处理 本体 语义相似度 贝叶斯估计 BETA分布 

分 类 号：TP391[自动化与计算机技术—计算机应用技术]