Research map / Large language models / Language, safety and interpretability
An Affect-Rich Neural Conversational Model with Biased Attention and Weighted Cross-Entropy Loss
Peixiang Zhong, Di Wang, Chunyan Miao
Read the paperCheck your idea against 137,217 papers
AI-extracted. The measured effects below were read by AI from the paper's full text (no summary yet). Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.
Measured effects reported in the paper
| What was changed | Where | Measured | Effect |
|---|---|---|---|
| Add local-importance affective attention and affective objective | Seq2Seq with attention · DailyDialog out-domain test | perplexity: 124.3 -> 113.7 | Improved |
| Add local-importance affective attention and affective objective | Seq2Seq with attention · Preference test | content preference: 66 (+3.1%) -> 80 (+25.0%) | Improved |
| Add local-importance affective attention and affective objective | Seq2Seq with attention · Preference test | emotion preference: 32 (+23.1%) -> 49 (+88.5%) | Improved |
| Add local-importance affective attention (ablation) | Seq2Seq with attention · MCT | emotion quality score: 0.712 -> 0.856 (+20.2%) | Improved |
| Add affective objective (ablation) | Seq2Seq with attention · MCT | emotion quality score: 0.712 -> 0.828 (+16.3%) | Improved |
| Combine local-importance affective attention with affective objective (ablation) | Seq2Seq with attention · MCT | emotion quality score: 0.712 -> 0.908 (+27.5%) | Improved |
| Replace uniform term importance with global term importance (ablation) | affective attention · MCT | emotion quality score: 0.832 -> 0.862 (+16.9% -> +21.1%) | Improved |
| Replace global term importance with local term importance (ablation) | affective attention · MCT | emotion quality score: 0.862 -> 0.856 | No clear effect |
| Increase affective loss coefficient δ (ablation) | S2S-AO · MCT test responses | distinct affect-rich words (VAD norm threshold 3): 36 -> 50 -> 69 | Improved |
| Increase affective attention coefficient γ beyond its small-value range (ablation) | affective attention · test responses | distinct affect-rich words | Hurt |
| Set γ in the range [20, ∞] (ablation) | affective attention · test responses | response length | Hurt |
| Increase affect embedding strength λ from 0 to 0.3 (ablation) | AR-S2S · test responses | distinct affect-rich words | Improved |
Closest papers by meaning
- AffectGPT: A New Dataset, Model, and Benchmark for Emotion Understanding with Multimodal Large Language Models (ICML 2025 · similarity 0.85)
- Nano-EmoX: Unifying Multimodal Emotional Intelligence from Perception to Empathy (CVPR 2026 · similarity 0.85)
- The Good, The Bad, and Why: Unveiling Emotions in Generative AI (ICML 2024 · similarity 0.85)
- Dual-View Learning for Conversational Emotion Recognition Through Context and Emotion-Shift Modeling (AAAI 2025 · similarity 0.84)
- MME-Emotion: A Holistic Evaluation Benchmark for Emotional Intelligence in Multimodal Large Language Models (ICLR 2026 · similarity 0.84)
- SD-Eval: A Benchmark Dataset for Spoken Dialogue Understanding Beyond Words (NeurIPS 2024 · similarity 0.84)
Writing a paper in this area? scime draft is a free LaTeX editor with AI included, real-time co-authors and a check of your references and venue rules before you submit. It also imports OpenAI Prism projects.
