Data provenance demonstrator

This application is used to demonstrate the feasibility and possible implementation of a recommendation system that is transparent in the way that it performs recommendations. This is done by providing users a way to view the data and algorithms used in producing a recommendation. The data used for this demonstrator is the Kaggle Netflix Prize dataset.

There are 2 modules in this application:

1. Recommender module

This module is responsible for producing recommendations, and contains some different algorithms used for recommendation. Each recommender in this module is responsible for producing its own provenance data, which is then consumed by the web module. The recommenders in this module can be customized, or a new one written as long as it conforms to the interfaces specified in this documentation.

2. Web module

Written using SpringBoot, this module is the web application that is used for demonstration purposes. The specific details of this module can be found in this documentation. The web application is self-contained and requires no setup to build and run.

Running the application

The application is buildable with Gradle and requires JDK 11 at least to build and run. The following command will build and run the application:

./gradlew bootRun

The application can be accessed in your browser at http://localhost:8080.

Using the application

The application can be used without the need to login, and the recommendations produced will contain provenance information when viewing the library. There are also three dummy accounts that have been pre-loaded with ratings to demonstrate provenance information for different users. The login details are as follows:

USERNAME:PASSWORD

jens:password
matthias:password
sam:password

An example of provenance data

-- Todo, insert the guide here for some interesting details

API specification

This documentation covers the specification of the API used by the web-server. Some endpoints are appended with .json as they share the same name as some pages of the web server.

POST /

Used for logging into the application. Currently, this endpoint doesn't return a different status for login failure, as the login failure is rendered server-side.

GET /library.json

GET /movie/{id}

GET /recommendations.json

GET /provenance

PUT /movie/rate

Additional information

There are several python scripts in the data-analysis directory that are used for creating SQL files and models that are used by the recommender module. The details of these scripts are in this documentation.

Credits

• Matthias Galster (University of Canterbury)
• Jens Dietrich (Victoria University of Wellington)
• Sam Shankland (University of Canterbury)