Research map / Vision-language and multimodal / Visual question answering
Learning to Align Question and Answer Utterances in Customer Service Conversation with Recurrent Pointer Networks
Shizhu He, Kang Liu, Weiting An
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 |
|---|---|---|---|
| use separate customer and server encoders on top of shared conversation encoder (Model-II vs Model-I) (ablation) | RPN (regression loss) · CS alignment overall | F1 (%): 82.75 -> 84.86 | Improved |
| use separate customer and server encoders (Model-II vs Model-I) (ablation) | RPN (classification loss) · CS alignment overall | F1 (%): 82.82 -> 83.49 | Improved |
| use regression loss instead of classification loss (ablation) | RPN Model-II · CS alignment One-to-Many | F1 (%): 76.38 -> 80.39 | Improved |
| use regression loss instead of classification loss (ablation) | RPN Model-I · CS alignment One-to-Many | F1 (%): 69.46 -> 78.37 | Improved |
| use regression loss instead of classification loss (ablation) | RPN Model-I · CS alignment overall | F1 (%): 82.82 -> 82.75 | No clear effect |
| use regression loss instead of classification loss (ablation) | RPN Model-II · CS alignment overall | F1 (%): 83.49 -> 84.86 | Improved |
| replace content matching with recurrent pointer network alignment | Q-A alignment · CS alignment overall | F1 (%): 49.28 -> 84.86 | Improved |
| replace pairwise two-stage method with RPN | SHCNN+CISIR · CS alignment overall | F1 (%): 79.5 -> 84.86 | Improved |
| use RPN instead of SHCNN+CISIR | SHCNN+CISIR · CS alignment None alignments | F1 (%): 83.32 -> 94.07 | Improved |
| add threshold strategy to content matching | Match-G (CNN) · CS alignment overall | F1 (%): 49.28 -> 46.49 | Hurt |
| remove conversation-encoder outputs from inputs of customer/server encoders (ablation) | RPN Model-II · CS alignment overall | F1 (%) | Hurt |
Closest papers by meaning
- Aligning Spoken Dialogue Models from User Interactions (ICML 2025 · similarity 0.85)
- Complex Sequential Question Answering: Towards Learning to Converse Over Linked Question Answer Pairs with a Knowledge Graph (AAAI 2018 · similarity 0.85)
- Addressee and Response Selection in Multi-Party Conversations With Speaker Interaction RNNs (AAAI 2018 · similarity 0.85)
- Building End-To-End Dialogue Systems Using Generative Hierarchical Neural Network Models (AAAI 2016 · similarity 0.84)
- Topic-Oriented Spoken Dialogue Summarization for Customer Service with Saliency-Aware Topic Modeling (AAAI 2021 · similarity 0.84)
- Teaching Language Models to Evolve with Users: Dynamic Profile Modeling for Personalized Alignment (NeurIPS 2025 · similarity 0.83)
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.
