A BiLSTM with attention and a gradient-reversal speaker-identity adversary reaches 93.2% accuracy and 94.2% F1 for depression detection on Androids-Corpus, a modest gain over its own non-adversarial baseline.
Italian bert and electra models,
1 Pith paper cite this work, alongside 12 external citations. Polarity classification is still indexing.
1
Pith paper citing it
12
external citations · OpenAlex
fields
cs.LG 1years
2026 1verdicts
CONDITIONAL 1representative citing papers
citing papers explorer
-
Multimodal Domain Generalization for Depression Detection: An Attention-Based BiLSTM Network with Domain-Adversarial Training
A BiLSTM with attention and a gradient-reversal speaker-identity adversary reaches 93.2% accuracy and 94.2% F1 for depression detection on Androids-Corpus, a modest gain over its own non-adversarial baseline.