Identify the specific SEAConv design elements responsible for its advantage

Identify which specific design elements within the Spatial Edge-Augmented Convolution (SEAConv) layer drive its performance advantage over alternative message-passing operators for football match-phase classification, and determine whether SEAConv's advantage over attention-based message passing generalizes to other tracking-data sets.

Background

The paper introduces SEAConv as a custom graph neural network message-passing layer for football tracking data. Unlike standard attention-based layers, SEAConv incorporates edge attributes—such as distances, relative angles, and team membership—both into attention coefficients and directly into the messages. In the reported experiments, SEAConv achieves the best macro F1 performance among the evaluated message-passing operators, although its margin over GATConv is small.

The authors explicitly leave unresolved which components of SEAConv account for its empirical advantage and whether that advantage is specific to the data set and match-phase task studied or extends to other sports-tracking data sets.