基于Fisher线性判别分析的情景感知推荐方法  

Context-aware recommendation method based on Fisher linear discriminant analysis

作　　者：杨茜[1] 

机构地区：[1]郑州大学体育学院,河南郑州450044

出　　处：《计算机工程与设计》2018年第3期848-853,共6页Computer Engineering and Design

基　　金：河南省教育厅科学技术研究重点基金项目(13B520253)

摘　　要：为解决现有推荐方法无法兼顾多种度量准则,提出一种基于线性判别分析的情景感知推荐方法。获取用户视图下的偏好项目特征、项目视图下的项目吸引度等多视图数据,通过特征融合、投影变换,在最佳鉴别矢量空间引入Fisher判别准则,采用Lagrange乘子法求解最优投影方向。实验结果表明,与现有方法相比,所提方法降低了时间开销,准确度平均提高18.91%,多样性平均提高32.79%,验证了其能够兼顾多种度量准则,提高了推荐质量。To solve the problem that the existing methods are incapable to cove d ifferent measure methods ＆ a context-awarerecommendation method based on Fisher linear discriminant analysis was proposed T he preference features from user^s v iew,and the item attraction from item^s view were g o t, Feature fusion and projection transformation were introduced, and Fisherdiscriminant criterion and Lagrange a lgorithm were used to compute the optimal vector space. Experimental extensive data set show that the proposed method can increase accuracy and diversity by about 18. 91by comparing wi th existing methods,wi th lower expenditure of t im e, which indicates th a t different measure methods and improve the accuracy of recommendation.

关 键 词：多视图学习 线性判别分析 FISHER准则 LAGRANGE乘子法 推荐系统 

分 类 号：TP391.3[自动化与计算机技术—计算机应用技术]