Основні поняття
EmotionICは、会話における感情認識のための感情慣性と伝染駆動依存モデリングを提案する。
Анотація
Abstract:
Emotion Recognition in Conversation (ERC) has gained attention due to human-computer interface advancements.
EmotionIC model integrates IMMHA, DiaGRU, and SkipCRF components for thorough conversation modeling.
Introduction:
ERC aims to identify emotions in conversations.
Contextual information is crucial for accurate emotion recognition.
Related work:
Various methods like LSTM, Transformer models have been proposed for ERC.
Existing methods struggle with capturing long-distance context dependencies effectively.
Our approach:
EmotionIC combines IMMHA and DiaGRU to capture global and local contextual information.
SkipCRF is introduced to model emotional flows at the classification level.
Experimental settings:
Evaluation metrics include Weighted-F1, Accuracy, Macro-F1, Micro-F1 on four benchmark datasets.
RoBERTa model is used for feature extraction with 1024 dimensions as input.
Results and analysis:
EmotionIC outperforms baseline models on all datasets.
Performance improvements attributed to thorough context modeling at both feature-extraction and classification levels.
Статистика
EmotionICはすべてのベンチマークデータセットでベースラインモデルを上回る結果を示しました。