Widespread machine learning methods behind 'link prediction' are performing very poorly, researchers find
Feb 2024, phys.org
Marketing vs Sciencing:
- Link prediction is a widespread machine learning task that evaluates the links in a network and tries to predict what the next links will be.
- New research establishes that the metric used to measure link prediction performance is missing crucial information, and link prediction tasks are performing significantly worse than popular literature indicates.
- They recommend that machine learning researchers stop using the standard practice metric for measuring link prediction ("area under curve" or AUC) and introduce a new metric.
- Authors discovered fundamental mathematical limitations to using low dimensional embeddings for link predictions, and that AUC can not measure these limitations, and conclude that AUC does not accurately measure link prediction performance.
- They recommend a new metric called VCMPR
via University of California Santa Cruz and University of Pennsylvania: Menand, Nicolas et al, Link prediction using low-dimensional node embeddings: the measurement problem, Proceedings of the National Academy of Sciences (2024). DOI: 10.1073/pnas.2312527121.
Image credit: AI Art - Room 404 - 2024
No comments:
Post a Comment