slope one 推荐算法0.pdfVIP

slope one 推荐算法 阿凡 /bicloud 内容 • slope one算法 • Slope one算法实例 • 分布式推荐算法 • Slope one in Mahout • Slope one @ 淘宝 • 参考资料 slope one算法 • Slope One 算法 是一种基于评分的预测算法, 本质上也是一种基于项目的算法。与一般的基于项目的算法不同, 该算法不计算 项目之间的相似度, 而是用一种简单的线性回归模型进行预测 （可以扩展） 算法易于实现, 计算速度快, 可扩展性好, 同时对数据稀疏性有较好的适应性 http://www.hitflip.de/dvds/title/Ozean_in_Fl ammen_1660/ Slope one 算法实例 Slope one 算法实例 • 经过上述处理，计算得到任意两个商品之间的平均差值 When done we have a table we can use to look up the average difference between any two items • This completes the preprocessing step • Empty cells contain inverses that are omitted here • 对称矩阵 Slope one 算法实例 •Let’s recommend an item for user X （假设现在给用户X推荐商品） •There are two potential candidates: item 103 and item 104 •We want to predict X’s preferences for both items and recommend the one user X would prefer •We need to do this using all of X’s existing items: 101,102 根据用户X评过分的商品101和102进行推荐 Slope one 算法实例 •Predict Xs preference for item 103 using item 101 •Look up the pre-computed average preference difference between items 103 and 101: -0.75 •Use this to predict X’s rating for item 103 based on item 101 X’s rating for item 101: 5 X’s predicted preference for item 103 using 101: -0.75 + 5 = 4.25 Slope one 算法实例 •Predict Xs preference for item 103 using item 102 •Look up the pre-computed average preference difference between items 103 and 102: -1.75 •Use this to predict X’s rating for item 103 based on item 102 •X’s rating for item 102: 3.5 •X’s predicted preference for item 103 using 102: -1.75 + 3.5 = 1.75 Slope one 算法实例 分布式推荐算法 为什么采用分布式: 1.Dataset does not fit into memory （数据无法直接加载进内存） 2.Need for processing exceeds what can be