摘要
从理论上分析了评分偏差对于推荐质量的影响;基于潜在偏好及已知评分对评分偏差进行度量,其中潜在偏好通过心理测量学模型计算得出;通过设定不同的评分偏差水平,对评分偏差的影响进行了实验验证.理论分析及实验验证表明:评分偏差可导致推荐准确度及覆盖度下降;基于高质量的评分数据,协同过滤算法可为用户作出好的推荐.
The effect of the rating residual on recommendation quality was analyzed. The rating residual was measured through user ratings and latent preferences. Latent preferences were computed with psychometric models. With different levels of rating residual, the effect of the rating residual was experimentally evaluated on real world datasets. Theoretical analysis and experimental results show that rating residual has negative effects on recommendation accuracy and coverage. Based on high quality of data, collaborative filtering algo- rithms can make precise recommendations for users.
出处
《北京航空航天大学学报》
EI
CAS
CSCD
北大核心
2012年第6期823-828,共6页
Journal of Beijing University of Aeronautics and Astronautics
基金
国家自然科学基金资助项目(61170189
60973105)
软件开发环境国家重点实验室自主研究课题资助项目(SKLSDE-2011ZX-03)
关键词
人工智能
信号过滤与预测
信息检索
评分偏差
数据质量
协同过滤
推荐准确度
覆盖度
artificial intelligence
signal filtering and prediction
information retrieval
rating residual
data quality
collaborative filtering
recommendation accuracy
coverage
