A random-walk pattern based graph learner that encodes semantic and anonymous paths and uses transformer attention to select task-relevant substructures, outperforming message passing GNNs on many benchmark tasks.
(3) Using a mean aggregator over encoded patterns instead of a transformer
1 Pith paper cite this work. Polarity classification is still indexing.
1
Pith paper citing it
fields
cs.LG 1years
2025 1verdicts
CONDITIONAL 1representative citing papers
citing papers explorer
-
Beyond Message Passing: Neural Graph Pattern Machine
A random-walk pattern based graph learner that encodes semantic and anonymous paths and uses transformer attention to select task-relevant substructures, outperforming message passing GNNs on many benchmark tasks.