Artificial Intelligence Recommendation：User特徵預取

情境介紹

擷取User特徵的方式有以下兩種：

• 阻塞擷取User特徵：因為User特徵會在召回或過濾階段用到， 必須先擷取User特徵，才能進行推薦流程。

• 非阻塞擷取User特徵：為了提高效能，需要非同步擷取User特徵， 之後才能在粗排或精排階段使用。

同時， User特徵的預取也需要AB實驗參數的支援。User特徵的擷取，在配置上是與特徵擷取相同的。

阻塞擷取User特徵

您可以利用UserFeatureConfs進行配置，UserFeatureConfs支援多情境配置，home_feed是配置的情境名稱，樣本內容如下。從下述樣本中可以看出當前配置和FeatureConfs配置是完全類似的。 UserFeatureConfs會在整個推薦流程之前執行，並且是阻塞式的，載入完User特徵之後才進行推薦流程。

{
  "UserFeatureConfs": {
    "home_feed": {
      "AsynLoadFeature": true,
      "FeatureLoadConfs": [
        {
          "FeatureDaoConf": {
            "AdapterType": "hologres",
            "HologresName": "pairec-holo",
            "FeatureKey": "user:uid",
            "UserFeatureKeyName": "client_str",
            "HologresTableName": "dwd_ali_user_all_feature_v2_holo",
            "UserSelectFields": "*",
            "FeatureStore": "user"
          },
          "Features": [
            {
              "FeatureType": "new_feature",
              "FeatureName": "day_h",
              "Normalizer": "hour_in_day",
              "FeatureStore": "user"
            },
            {
              "FeatureType": "new_feature",
              "FeatureName": "week_day",
              "Normalizer": "weekday",
              "FeatureStore": "user"
            }
          ]
        }
      ]
    }
  }
}

其中AsynLoadFeature表示是否非同步載入User特徵。當有多個FeatureLoadConfs時，可以使用非同步並發方式進行載入。其他參數配置說明，請參見配置欄位說明。

非阻塞擷取User特徵

如果您想擷取User特徵而不阻塞推薦流程，可以添加FeatureAsyncLoad並將其配置為true，樣本內容如下：

{
  "UserFeatureConfs": {
    "home_feed": {
      "AsynLoadFeature": true,
      "FeatureLoadConfs": [
        {
          "FeatureDaoConf": {
            "AdapterType": "hologres",
            "HologresName": "pairec-holo",
            "FeatureKey": "user:uid",
            "UserFeatureKeyName": "client_str",
            "HologresTableName": "dwd_ali_user_all_feature_v2_holo",
            "UserSelectFields": "*",
            "FeatureStore": "user"
          },
          "Features": [
            {
              "FeatureType": "new_feature",
              "FeatureName": "day_h",
              "Normalizer": "hour_in_day",
              "FeatureStore": "user"
            },
            {
              "FeatureType": "new_feature",
              "FeatureName": "week_day",
              "Normalizer": "weekday",
              "FeatureStore": "user"
            }
          ]
        },
        {
          "FeatureDaoConf": {
              "AdapterType": "hologres",
            "FeatureAsyncLoad": true,
            "HologresName": "pairec-holo",
            "FeatureKey": "user:uid",
            "UserFeatureKeyName": "client_str",
            "HologresTableName": "dwd_ali_user_all_feature_v3_holo",
            "UserSelectFields": "*",
            "FeatureStore": "user"
          },
          "Features": []
        }
      ]
    }
  }
}

上述檔案配置擷取兩類特徵：

• 從dwd_ali_user_all_feature_v2_holo中擷取所有特徵，該內容未配置FeatureAsyncLoad，表示這些特徵是阻塞式擷取的。

• 從dwd_ali_user_all_feature_v3_holo擷取所有特徵。該內容配置了FeatureAsyncLoad=true，則表示這些特徵使用非同步方式擷取。

其他參數配置說明，請參見配置欄位說明。

預取User特徵並放入Cache

為提高系統效能，建議您在執行推薦流程之前，採用非同步方式擷取User特徵，並將這些特徵暫時儲存。待實際需求時，再將緩衝的特徵放到使用者特徵集中，確保高效利用。樣本內容如下：

{
  "UserFeatureConfs": {
    "home_feed": {
      "AsynLoadFeature": true,
      "FeatureLoadConfs": [
        {
          "FeatureDaoConf": {
            "AdapterType": "hologres",
            "HologresName": "pairec-holo",
            "FeatureKey": "user:uid",
            "UserFeatureKeyName": "client_str",
            "FeatureAsyncLoad": true,
            "HologresTableName": "dwd_ali_user_all_feature_v2_holo",
            "UserSelectFields": "*",
            "CacheFeaturesName" :"test",//cache map
            "FeatureStore": "user"
          },
          "Features": [
            
            {
              "FeatureType": "new_feature",
              "FeatureName": "day_h",
              "Normalizer": "hour_in_day",
              "FeatureStore": "user"
            },
            {
              "FeatureType": "new_feature",
              "FeatureName": "week_day",
              "Normalizer": "weekday",
              "FeatureStore": "user"
            }
          ]
        }
      ]
    }
    }
}

其中關鍵配置說明如下，其他參數配置說明，請參見配置欄位說明。

• "FeatureAsyncLoad" = true：表示使用非同步方式擷取User特徵。

• "CacheFeaturesName": "test"：表示擷取到的特徵會緩衝到名稱為test的Cache Map中，而不是真正放入到User特徵中。後面用到特徵時，才需要引用。每組特徵都可以使用不同的CacheFeaturesName名稱。

引用User的Cache特徵

在粗排或者精排階段引用預先載入的User特徵時，引用的特徵需要在FeatureConfs裡配置。樣本內容如下：

{
    "FeatureConfs":{
        "home_feed_rebuild_v22":{
            "AsynLoadFeature":false,
            "FeatureLoadConfs":[
                {
                    "FeatureDaoConf":{
                        "LoadFromCacheFeaturesName":"test",
                        "FeatureStore":"user"
                    }
                },
                {
                    "FeatureDaoConf":{
                        "LoadFromCacheFeaturesName":"test1",
                        "FeatureStore":"user"
                    }
                }
            ]
        }
    }
}

其中關鍵配置說明如下，其他參數配置說明，請參見配置欄位說明。

• 引用特徵不需要並發的擷取，設定"AsynLoadFeature"= false即可。

• LoadFromCacheFeaturesName：表示從哪個Cache Map中擷取User特徵。名稱與CacheFeaturesName保持一致。

配置欄位說明

FeatureLoadConfs/FeatureDaoConf中的配置欄位說明如下：