基于置信度加权的单类协同过滤推荐算法

One-class collaborative filtering based on confidence weighting

针对推荐系统中单类协同过滤(OCCF)可解释性差、数据噪声多的缺陷,提出了一种基于置信度加权的单类协同过滤推荐算法。算法通过置信度函数将用户隐性反馈映射为置信概率,并将该函数集成到隐性反馈推荐模型(IFRM)框架中,形成了隐性反馈置信度加权推荐模型(CWIFRM);在此基础上,针对CWIFRM基于随机梯度下降提出了异构置信度优化算法。实验结果表明,该模型在多个数据集上都具有更好的推荐效果,异构置信度优化算法使推荐质量得到了进一步提高,验证了CWIFRM具有较强的适用性、可解释性和抗噪声能力。

Aiming at addressing the problem that implicit feedback has poor interpretability and high data noise in recommender system,this paper proposed a new one-class collaborative filtering algorithm based on confidence weighting. In order to map user＇s implicit feedback to confidence probability,the algorithm built confidence function and integrated the function into IFRM framework to form the confidence-weighted implicit feedback recommendation model（ CWIFRM）. Furthermore,the paper proposed a heterogeneous confidence optimization algorithm based on stochastic gradient descent for CWIFRM. Experiment shows that the proposed model has a better performance on multiple datasets than the other four algorithms. In addition,the heterogeneous confidence optimization algorithm has further improved the quality of recommendation. As a result,it demonstrates that CWIFRM has strong applicability,interpretability and anti-noise ability.