文献资源主题向量表征方法分析及改进研究  

On Literature Resources Theme Vector Representation Method and Its Improvement

作　　者：汪英姿[1] 徐飞[1] 

机构地区：[1]常州大学,常州213164

出　　处：《情报杂志》2012年第11期141-144,共4页Journal of Intelligence

摘　　要：DDM模型在现在流行的LDA模型中引入了残差分量来弥补判别信息的丢失,而由于文档词汇服从幂律分布,导致残差分量会受到高频主题词汇的影响。本文提出了HDDM模型,利用改进的TF-IDF函数,设置残差分量采样时的权重,提高了低频词汇在残差分量中的作用。实验证明这种模型提高了个性化推荐的推荐质量,使得查准率得到进一步的提升。DDM model introduced into the currently popular LDA model the residual component to make up for the loss of the discriminated information.Because the vocabulary of document obeys the power-law distribution,residual component will be affected by the high frequency theme vocabulary.This paper puts forward HDDM model,using the improved TF-IDF function,sets the sampling weight of the residual component,and improves the low frequency vocabulary＇s function in residual component.Experiments show that this model improves the quality of personalized recommendation,further promoting the precision.

关 键 词：个性化推荐 DDM模型 HDDM模型 TF-IDF函数 

分 类 号：TP31[自动化与计算机技术—计算机软件与理论]