實用推薦系統(tǒng)

第1部分　推薦系統(tǒng)的準備工作
第1章　什么是推薦 ..................................................................................3
1.1　現實生活中的推薦 ............................................................................3
1.1.1　推薦系統(tǒng)在互聯(lián)網上大顯身手 .......................................................5
1.1.2　長尾 ..............................................................................................5
1.1.3 Netflix 的推薦系統(tǒng) ...........................................................................6
1.1.4　推薦系統(tǒng)的定義 .............................................................................13
1.2　推薦系統(tǒng)的分類 .................................................................................15
1.2.1　域 .....................................................................................................16
1.2.2　目的 .................................................................................................16
1.2.3　上下文 .............................................................................................17
1.2.4　個性化級別 .....................................................................................17
1.2.5　專家意見 .........................................................................................19
1.2.6　隱私與可信度 .................................................................................19
1.2.7　接口 .................................................................................................20
1.2.8　算法 .................................................................................................23
1.3　機器學習與Netflix Prize .............................................................................24
1.4 MovieGEEKs網站 .......................................................................................25
1.4.1　設計與規(guī)范 .....................................................................................27
1.4.2　架構 .................................................................................................27
1.5　構建一個推薦系統(tǒng) .....................................................................................29
小結 ......................................................................................................................31
第2章　用戶行為以及如何收集用戶行為數據 ..................................................32
2.1　在瀏覽網站時Netflix如何收集證據 ..........................................................33
2.1.1 Netflix 收集的證據 .........................................................................35
2.2　尋找有用的用戶行為 .................................................................................37
2.2.1　捕獲訪客印象 .................................................................................38
2.2.2　可以從瀏覽者身上學到什么 .........................................................38
2.2.3　購買行為 .........................................................................................43
2.2.4　消費商品 .........................................................................................44
2.2.5　訪客評分 .........................................................................................45
2.2.6　以（舊的）Netflix 方式了解你的用戶 .........................................48
2.3　識別用戶 .....................................................................................................49
2.4　從其他途徑獲取訪客數據 .........................................................................50
2.5　收集器 .........................................................................................................50
2.5.1　構建項目文件 .................................................................................52
2.5.2　數據模型 .........................................................................................52
2.5.3　告密者（snitch）：客戶端證據收集器 ..........................................53
2.5.4　將收集器集成到 MovieGEEKs 中 .................................................54
2.6　系統(tǒng)中的用戶是誰以及如何對其進行建模 .............................................57
小結 ......................................................................................................................60
第3章　監(jiān)控系統(tǒng).........................................................................................61
3.1　為什么添加儀表盤是個好主意 .................................................................62
3.1.1　回答“我們做得怎么樣？” ...........................................................62
3.2　執(zhí)行分析 .....................................................................................................64
3.2.1　網站分析 .........................................................................................64
3.2.2　基本統(tǒng)計數據 .................................................................................64
3.2.3　轉化 .................................................................................................65
3.2.4　分析轉化路徑 .................................................................................69
3.2.5　轉化路徑 .........................................................................................70
3.3　角色 .............................................................................................................73
3.4 MovieGEEKs儀表盤 ...................................................................................76
3.4.1　自動生成日志數據 .........................................................................76
3.4.2　分析儀表盤的規(guī)范和設計 .............................................................77
3.4.3　分析儀表盤示意圖 .........................................................................77
3.4.4　架構 .................................................................................................78
小結 ......................................................................................................................81
第4章　評分及其計算方法............................................................................82
4.1　用戶-商品喜好 ...........................................................................................83
4.1.1　什么是評分 .....................................................................................83
4.1.2　用戶 - 商品矩陣 .............................................................................84
4.2　顯式評分和隱式評分 .................................................................................86
4.2.1　如何選擇可靠的推薦來源 .............................................................87
4.3　重溫顯式評分 .............................................................................................88
4.4　什么是隱式評分 .........................................................................................88
4.4.1　與人相關的推薦 .............................................................................90
4.4.2　關于計算評分的思考 .....................................................................90
4.5　計算隱式評分 .............................................................................................93
4.5.1　看看行為數據 .................................................................................94
4.5.2　一個有關機器學習的問題 .............................................................98
4.6　如何計算隱式評分 .....................................................................................99
4.6.1　添加時間因素 ...............................................................................102
4.7　低頻商品更有價值 ...................................................................................105
小結 ...............................................................