Skip to content

Latest commit

 

History

History
38 lines (31 loc) · 2.63 KB

README.md

File metadata and controls

38 lines (31 loc) · 2.63 KB

Online Service for MovieLens Recommender

The overall online engineering framework is developed based on SpringBoot+K8S, and it is also convenient for everyone to deploy to their own servers. We can configure our developing environment according to the installation instructions below. In this demo, our online pipeline consists of several modules such as user modeling, recall, ranking, diversity and summary retrieval, as described in the figure below. When we follow the instructions to finish the last step, an end to end movie recommendation system will be presented completely. If you are Chinese developer, you may like to visit our CN Doc.

image

1. Install online-serving components

You need to maven install online-serving components before launching our online recommend service.

  1. MetaSpore Serving is used to do model inference.
  2. feature-extract is a maven plugin which generate domains and repositories of MongoDB using its table yaml files. More information
  3. experiment-pipeline is an experiment framework which help us do A/B testing more easily. More information.
cd MetaSpore/java/online-serving
mvn clean install

2. Create application-dev.properties

You need to create resources/application-dev.properties from application-template.properties and specify:

  1. MongoDB related configurations.
  2. MetaSpore Serving related configurations.
  3. Milvus related configurations.
  4. (Just ignore mysql configurations. We don't use it in this demo but we support multi data sources.)

3. Install and configure Consul

You could modify Consul's Key/Value pair to dynamically change the online A/B testing strategies.

  1. Install Consul and launch it.
  2. Visit Consul's portal, and create a new Key/Value pair:
    1. Key is config/test/scene-config
    2. Copy this YAML content as value.
    3. The config file of Consul of our demo is here.

4. Launch recommend online service

You could run online service entry point (MovielensRecommendApplication.java) and test it now. For example: curl http://localhost:8080/user/10 to get recommended movies for user whose userId is equal to 10.