基于Bhattacharyya系数的改进相似度度量方法  

Research of improving similarity measure based on Bhattacharyya coefficient

作　　者：杜茂康[1] 王忠思 宋强[2] DU Maokang;WANG Zhongsi;SONG Qiang(Key Laboratory of Electronic Commerce and Logistics,Chongqing University of Posts and Telecommunications, Chongqing 400065,P.R.China;Chongqing Communication Administration,Chongqing 401121,P.R.China)

机构地区：[1]重庆邮电大学电子商务与现代物流重点实验室,重庆400065 [2]重庆市通信管理局,重庆401121

出　　处：《重庆邮电大学学报（自然科学版）》2018年第5期699-704,共6页Journal of Chongqing University of Posts and Telecommunications(Natural Science Edition)

基　　金：重庆市社科规划项目(K2015-59);重庆市前沿与应用基础研究项目(cstc2015jcyjA10081);重庆市社会科学规划管理项目(2015SKZ09)~~

摘　　要：相似度度量是基于邻居的协同过滤推荐算法中的关键步骤,对推荐结果的优劣有至关重要的影响。基于Bhattacharyya系数的相似度度量方法虽然解决了依赖于共同评分的问题,但忽略了评分值绝对数量对结果的影响。同时,当项目间相同评分值数量占比小时,基于Bhattacharyya系数的相似度度量方法存在计算准确性差的缺点。为此,引入Laplace校准法和权重赋值法对该相似度度量方法进行改进。改进后的方法不仅克服了原方法的不足,而且还充分利用所有评分信息,提升计算的准确性。数据实验结果表明,提出的相似度度量方法性能优于改进前的算法及传统的度量方法。Similarity measure is the key passage in the collaborative filtering algorithm based on neighbors,which will have an important effect on the prediction result.The similarity measure based on the Bhattacharyya coefficient resolves the problem of dependence on the corated items in traditional measure methods.However,it ignores the influence of the absolute amount to ratings.Besides,while the ratio of common ratings is rather low,the similarity measure based on the Bhattacharyya coefficient performs poorly in term of precise.Therefore,this paper introduces Laplace Calibration method and weight to improve the similarity measure.The improved measure not only overcomes the shortcomings of the original measures,but utilizes all the value of ratings sufficiently.Then the accuracy of calculation is enhanced.The result of experiment reveals that the similarity measure improved in this paper has the better performance than the original measure and the traditional methods.

关 键 词：相似度度量 Laplace校准法 权重法 BHATTACHARYYA系数 

分 类 号：TP391[自动化与计算机技术—计算机应用技术]