Multi-View Gated Graph Attention Network achieves 90% accuracy on ADReSSo

A Multi-View Gated Graph Attention Network has been proposed to detect Alzheimer's Disease by analyzing spontaneous speech through a "content-structure-flow" framework. The system transcribes audio via Automatic Speech Recognition (ASR) to construct semantic, dependency, and co-occurrence graphs, utilizing Pointwise Mutual Information (PMI) from a normative corpus to quantify narrative logic.

• The model integrates these distinct graph views using an adaptive gated fusion mechanism to address clinical heterogeneity.
• Evaluated on the ADReSSo dataset, the approach achieves 90.00% accuracy.
• Ablation studies confirm that both the PMI-based graph and the gating mechanism are essential for robust classification.

The authors consider this significant because it addresses non-linear structural disruptions in pathological language that many existing systems overlook. The source code is publicly.