基于评分预测与排序预测的协同过滤推荐算法

协同过滤推荐算法在电子商务领域运用广泛.之前的研究要么仅从评分预测的角度来研究，要么仅从排序预测的角度来研究.为了兼顾这两个方面，本文在传统的基于评分预测的PMF（Probabilistic Matrix Factorization）算法和基于排序预测的xCLiMF（Extended Collaborative Less-is-More Filtering）算法的基础上提出了一种基于评分预测与排序预测的协同过滤推荐算法URA（Unified Recommendation Algorithm），该方法通过在PMF和xCLiMF算法中共享用户和推荐对象的特征空间，利用PMF算法来学习高精度的用户和推荐对象的特征向量，从而进一步增强排序推荐性能.实验验证，该方法在评价指标NDCG和ERR下均优于PMF和xCLiMF算法，且复杂度与评分点个数线性相关.URA算法可运用于互联网信息推荐领域的大数据处理.

Collaborative Filtering Recommendation Algorithm Based on Rating Prediction and Ranking Prediction

Collaborative filtering (CF) recommendation algorithm is widely used in the field of e-commerce.The previous researches on CF focused on either raring prediction or ranking prediction.In order to take into account these two aspects,a collaborative filtering recommendation algorithm based on rating prediction and ranking prediction (Unified Recommendation Algorithm,URA) is proposed.URA shares common latent features of users and items in PMF (Probabilistic Matrix Factorization,rating-oriented) and xCLiMF (Extended Collaborative Less-is-More Filtering,ranking-oriented) algorithms,and PMF learns improved latent features of users and items in URA,so that URA improves the performance of ranking recommendation.Experimental results showed that our proposed URA Algorithm outperformed PMF and xCLiMF algorithms over evaluation metrics NDCG and ERR,and that the complexity of URA is shown to be linear with the number of observed ratings.URA is suitable for big data processing in the field of internet information recommendation.