/quantum-link-prediction

Working repository for the science4cast competition using quantum walks

Primary LanguageJupyter Notebook

Link prediction using simple network metrics

Working repository for the science4cast competition.

In the notebook full_code.ipynb there is all the code to replicate the results of the submissions of team Bacalhink. The data to run the algorithm can be found in the science4cast website https://www.iarai.ac.at/science4cast/.

These methods have been initially inspired by a relaxation of a quantum walker L3 and L2 methods but then departed in favour of other metrics like PA. More details of the approach used can be found in the scientific report.

Competition submissions

Here is what we have submitted to the leaderboard:

  • Bacalhau à Brás: the PA method - performance: 0.89715
  • Bacalhau à Gomes de Sá: the AA method - performance: 0.87091
  • Bacalhau com Todos: combines AA and PA - performance: 0.91385
  • Bacalhau com Natas: PA with time weights - performacne: 0.90364
  • Bacalhau à Lagareiro: combines AA and PA with time weights - performance: 0.91853

Free parameters:

  • a: defines the balance in the linear combination of AA with PA
  • (θ0, θ1, θ2, θ3): defines the time-weighting function