交替最小二乘ALS（Alternating Least Squares）是矩陣分解的一種演算法，常用於推薦系統中，尤其是協同過濾情境。其主要目標是將一個使用者-物品評分矩陣分解為兩個低階矩陣的乘積，從而實現降維、填補缺失值和發現潛在的使用者偏好和物品特徵。
支援的計算資源
MaxCompute/Flink
輸入/輸出
輸入樁
輸入的上遊組件支援：
輸出樁
輸出的User因子和Item因子對應下遊組件：ALS評分
配置組件
在Designer工作流程頁面添加ALS矩陣分解組件，並在介面右側配置相關參數：
參數類型 | 參數 | 描述 |
欄位設定 | user列名 | 輸入資料來源中，使用者ID列的名稱。該列資料必須是BIGINT類型。 |
item列名 | 輸入資料來源中，item項的列名。該列資料必須是BIGINT類型。 | |
打分列名 | 輸入資料來源中，使用者對item項的打分所在的列名。該列資料必須是數值型。 | |
參數設定 | 因子數 | 預設值為10，取值範圍為(0,+∞)。 |
迭代數 | 預設值為10，取值範圍為(0,+∞)。 | |
正則化係數 | 預設值為0.1，取值範圍為(0,+∞)。 | |
複選框 | 是否採用隱式偏好模型。 | |
隱式偏好係數 | 預設值為40，取值範圍為(0,+∞)。 | |
輸出表生命週期 | 輸出模型表的生命週期，單位天。 | |
執行調優 | 節點個數 | 取值範圍為1~9999。 |
單個節點記憶體大小 | 取值範圍為1024 MB~64*1024 MB。 |
使用樣本
使用以下資料作為ALS演算法模板的輸入資料，可以獲得輸出的user因子和item因子：
輸入資料來源
user_id
item_id
rating
10944750
13451
0
10944751
13452
1
10944752
13453
2
10944753
13454
2
10944754
13455
4
... ...
... ...
... ...
輸出的user因子表
user_id
factors
8528750
[0.026986524,0.03350178,0.03532385,0.019542359,0.020429865,0.02046867,0.022253247,0.027391396,0.018985065,0.04889483]
282500
[0.116156064,0.07193632,0.090851225,0.017075706,0.025412979,0.047022138,0.12534861,0.05869226,0.11170533,0.1640192]
4895250
[0.038429666,0.061858658,0.04236993,0.055866677,0.031814687,0.0417443,0.012085311,0.0379342,0.10767074,0.028392972]
... ...
... ...
輸出的item因子表
item_id
factors
24601
[0.0063337763,0.026349949,0.0064828005,0.01734504,0.022049638,0.0059205987,0.008568814,0.0015981696,0.0,0.013601779]
26699
[0.0027524426,0.0043066847,0.0031336215,0.00269448,0.0022347474,0.0020477585,0.0027995422,0.0025390312,0.0033011117,0.003957773]
20751
[0.03902271,0.050952066,0.032981463,0.03862796,0.048720762,0.027976315,0.02721664,0.018149626,0.0149896275,0.026251089]
... ...
... ...