Attribute-Enhanced Similarity Ranking for Sparse Link Prediction

João Mattos, Zexi Huang, Mert Kosan, Ambuj Singh, Arlei Silva · 2024

Read Attribute-Enhanced Similarity Ranking for Sparse Link Prediction by João Mattos, Zexi Huang, Mert Kosan, Ambuj Singh, Arlei Silva (2024) in the Tower of Babel, a free online library of public-domain works preserved in clean, searchable text. No sign-up. The full text is served as a plain-text file at a predictable URL, where text is present: https://antarctic-labs.com/tower-of-babel/arxiv-2412-00261.txt.

Description

João Mattos, Zexi Huang, Mert Kosan, Ambuj Singh, and Arlei Silva challenged graph neural networks on link prediction in twenty twenty-four. Link prediction guesses which missing connections a network will form. Neural methods report strong scores, but they train and test on balanced samples, while real networks are extremely sparse.

The authors show those scores do not carry over to the realistic imbalanced setting, where simple topology-based heuristics often win. Their method, Gelato, boosts a topological heuristic with node attributes learned from the graph. A ranking loss handles the imbalance, and a partition-based sampler picks hard negative examples efficiently. Experiments show Gelato beats the neural alternatives it was tested against. The paper urges more honest evaluation of link prediction under true sparsity.

Catalog record

CREATORJoão Mattos, Zexi Huang, Mert Kosan, Ambuj Singh, Arlei Silva
YEAR2024
COLLECTIONPAPERS
FORMATTXT
FILE SIZE70 KB
VERSION1.0
LICENSEhttp://creativecommons.org/licenses/by/4.0/
RIGHTSCC BY
SOURCEarXiv

Source: https://arxiv.org/abs/2412-00261

Browse the full library · About the Tower of Babel