Research map / Architectures and efficiency / Attention and transformers
Kalman Filtering Attention for User Behavior Modeling in CTR Prediction
Hu Liu, Jing LU, Xiwei Zhao, Sulong Xu, Hao Peng, Yutong Liu, Zehua Zhang, Jian Li, Junsheng Jin, Yongjun Bao, Weipeng Yan
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 |
|---|---|---|---|
| Replace Transformer attention with KFAtt-freq | Transformer behavior model · Amazon Electronics (All) | AUC: 0.8720 -> 0.8789 | Improved |
| Replace Transformer attention with KFAtt-freq | Transformer behavior model · Amazon Electronics (New queries) | AUC: 0.8488 -> 0.8578 | Improved |
| Plug KFAtt-freq into vanilla attention | Vanilla Attention · Amazon Electronics (All) | AUC: 0.8034 -> 0.8481 | Improved |
| Plug KFAtt-freq into DIN | DIN · Amazon Electronics (All) | AUC: 0.8317 -> 0.8524 | Improved |
| Add global query prior (KFAtt-base) to Transformer | Transformer · Amazon Electronics (All) | AUC: 0.8720 -> 0.8766 | Improved |
| Add frequency capping on top of KFAtt-base (KFAtt-freq) | KFAtt-trans · Amazon Electronics (All) | AUC: 0.8766 -> 0.8789 | Improved |
| Learned query-specific prior (sigma_q) instead of naive fixed prior sigma_q=1 (ablation) | KFAtt-base · Amazon Electronics (All) | AUC: 0.8740 -> 0.8766 | Improved |
| Include random error term in frequency capping rather than naive capping (ablation) | KFAtt-freq · Amazon Electronics (All) | AUC: 0.8754 -> 0.8789 | Improved |
| Coarser (2nd/1st-level category) deduplication instead of 3rd-level (ablation) | KFAtt-freq · Amazon Electronics (All) | AUC: 0.8789 -> 0.8775 / 0.8766 | Hurt |
| Coarser (2nd/1st-level category) deduplication instead of 3rd-level (ablation) | KFAtt-freq · Amazon Electronics (Infrequent queries) | AUC: 0.8496 -> 0.8504 / 0.8506 | Improved |
| Deploy KFAtt-freq online vs sum-pooling base model | JD.com ranking system · JD.com online A/B test | AUC: 4.4% CTR gain | Improved |
Closest papers by meaning
- From Feature Interaction to Feature Generation: A Generative Paradigm of CTR Prediction Models (ICML 2025 · similarity 0.85)
- AT4CTR: Auxiliary Match Tasks for Enhancing Click-Through Rate Prediction (AAAI 2024 · similarity 0.84)
- Listwise Preference Diffusion Optimization for User Behavior Trajectories Prediction (NeurIPS 2025 · similarity 0.84)
- Addressing Cold-Start Problem in Click-Through Rate Prediction via Supervised Diffusion Modeling (AAAI 2025 · similarity 0.83)
- Long-Sequence Recommendation Models Need Decoupled Embeddings (ICLR 2025 · similarity 0.82)
- UxSID: Semantic-Aware User Interests Modeling for Ultra-Long Sequence (NeurIPS 2026 · similarity 0.82)
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.
