All Products
Search
Document Center

Artificial Intelligence Recommendation:Engine configuration

Last Updated:Dec 10, 2024

Overall configuration overview

To simplify the use of the PAI-Rec recommendation engine, configurations of all stages involved in a recommendation process including recall, filtering, ranking, and re-ranking in configuration files are described in a comprehensive and detailed manner. Users do not need to write code. They can use configuration files to configure a whole recommendation process, including the storage service from which candidate items can be obtained, exposure blocking, ranking models, and re-ranking. If algorithm engineers and strategy engineers need to conduct multiple types of experiments, they can use the A/B testing system of the PAI-Rec engine.

Multiple storage services are available on Alibaba Cloud. PAI-Rec allows users to store data in Hologres, Tablestore, and ApsaraDB for Redis. Users can use different storage services based on their business requirements. More details about storage services are described in specific topics.

When many new items need to be recommended every day, users can configure a cold start solution.

PAI-Rec supports hot reloading. Therefore, configurations can be modified and republished by using the hot reloading feature.

The following code block shows the overall configuration of the PAI-Rec engine:

{
  "RunMode": "",
  "ListenConf": {},
  "RecallConfs": [],
  "SceneConfs": {},
  "FilterConfs": [],
  "FilterNames": {},
  "BeFilterConfs": {},
  "GeneralRankConfs": {},
  "AlgoConfs": [],
  "RankConf": {},
  "ColdStartGeneralRankConfs": {},
  "ColdStartRankConfs": {},
  "SortConfs": {},
  "DPPConf": {},
  "SortNames": {},
  "FeatureConfs": {},
  "UserFeatureConfs": {},
  "HologresConfs": {},
  "KafkaConfs": {},
  "RedisConfs": {},
  "MysqlConfs": {},        
  "GraphConfs": {},
  "FeatureStoreConfs": {},
  "KafkaConfs": {},
  "SlsConfs": {},
  "DatahubConfs": {},
  "BEConfs": {},
  "TableStoreConfs": {},
  "PrometheusConfig": {},
  "LogConf": {},
  "DebugConfs": {},
  "CallBackConfs": {},
  "PipelineConfs": {},
  "UserDefineConfs": {}
}

Parameter

Description

Support for hot reloading

RunMode

The runtime environment of the PAI-Rec engine. Valid values:

● daily: the daily environment. ● prepub: the staging environment. ● product: the production environment.

No

ListenConf

The listener configuration.

Yes

RecallConfs

The recall configuration.

Yes

SceneConfs

The scenario configuration. You can use specific recalls configured in the RecallConfs section for different scenarios.

Yes

FilterConfs

The filter configuration.

Yes

FilterNames

The custom names of filters. You can use specific filters configured in the FilterConfs section for different scenarios.

Yes

BeFilterConfs

Yes

AlgoConfs

The configuration of the coarse ranking model.

Yes

GeneralRankConfs

You can use specific coarse ranking models configured in the AlgoConfs section for different scenarios.

Yes

RankConf

You can use specific fine ranking models configured in the AlgoConfs section for different scenarios.

Yes

ColdStartGeneralRankConfs

You can use specific coarse ranking models for cold start configured in the AlgoConfs section for different scenarios.

Yes

ColdStartRankConfs

You can use specific cold start models configured in the AlgoConfs section for different scenarios.

Yes

SortConfs

The re-ranking configuration.

Yes

DPPConf

The diversification configuration.

Yes

SortNames

You can use specific re-ranking models configured in the SortConfs section for different scenarios.

Yes

FeatureConfs

You can load specific features for different scenarios.

Yes

UserFeatureConfs

You can load specific user features for different scenarios.

Yes

HologresConfs

The configuration of the connection between PAI-Rec and Hologres.

No

KafkaConfs

The configuration of the connection between PAI-Rec and ApsaraMQ for Kafka.

No

RedisConfs

The configuration of the connection between PAI-Rec and ApsaraDB for Redis.

No

MysqlConfs

The configuration of the connection between PAI-Rec and ApsaraDB RDS for MySQL.

No

GraphConfs

The configuration of the connection between PAI-Rec and GraphCompute.

No

FeatureStoreConfs

The configuration of the connection between PAI-Rec and FeatureStore.

No

SlsConfs

The configuration of the connection between PAI-Rec and Simple Log Service.

No

DatahubConfs

The configuration of the connection between PAI-Rec and DataHub.

No

BEConfs

No

TableStoreConfs

The configuration of the connection between PAI-Rec and Tablestore.

No

PrometheusConfig

The configuration of the connection between PAI-Rec and Managed Service for Prometheus.

No

LogConf

The log output configuration.

Yes

DebugConfs

The debugging configuration.

Yes

CallBackConfs

The callback configuration. For more information, see Callback interface.

Yes

PipelineConfs

The pipeline configuration.

Yes

UserDefineConfs

The user-defined configuration.

No