摘要
Apache Mahout推荐系统的评估方法目前仍局限在较传统的平均数绝对误差,不能真实反映推荐系统的准确率。故提出一种新的评估方法:使用M系数对Apache Mahout推荐系统的准确率和覆盖率进行综合评估。通过实验验证,发现平均数绝对误差越低并非准确率就越高,而最好的准确率和覆盖率平衡的结果是高覆盖率加上低相似度阈值。这种评估方法找到了Mahout作为个性化推荐系统的最佳实践。
The evaluation of Apache Mahout recommendation system is still limited in the traditional mean absolute error evaluation method,which cannot represent the accuracy of Apache Mahout recommendation system. Hence, a new evaluation method is proposed, that is, comprehensively evaluating the accuracy and coverage of Apache recommender algorithm using M coefficients. Through experi- ment, it is found that the lower mean absolute error does not imply the higher accuracy, whereas the optimistically compromised result is a high coverage and a low similarity threshold. This method is the best practice of Mahout as a personalized recommendation system.
出处
《济南大学学报（自然科学版）》
CAS
北大核心
2016年第1期47-50,共4页
Journal of University of Jinan(Science and Technology)
基金
国家自然科学基金(61170176)
关键词
协同过滤算法
算法评估
M系数
collateral filtering algorithm
algorithm evaluation
M coefficient
