Research map / Label-efficient and robust learning / Recognition and long-tail
Feature Interaction Interpretability: A Case for Explaining Ad-Recommendation Systems via Neural Interaction Detection
Michael Tsang, Dehua Cheng, Hanpeng Liu, Xue Feng, Eric Zhou, Yan Liu
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 |
|---|---|---|---|
| Encode AutoInt's top-40 detected interactions in DeepFM | DeepFM · Criteo | AUC: 0.8079 ±3e-4 -> 0.8097± 2e-4 | Improved |
| Encode AutoInt's top-40 detected interactions in DeepFM | DeepFM · Criteo | logloss: 0.4436±2e-4 -> 0.4420±2e-4 | Improved |
| Encode AutoInt's top-10 detected interactions in DeepFM | DeepFM · Avazu | AUC: 0.7792±3e-4 -> 0.7795±2e-4 | Improved |
| Encode AutoInt's top-10 detected interactions in DeepFM | DeepFM · Avazu | logloss: 0.3804± 9e-5 -> 0.3802±2e-4 | Improved |
| Encode AutoInt's top-40 detected interactions in AutoInt | AutoInt · Criteo | AUC: 0.8083 -> 0.8090±{2e-4} | Improved |
| Encode AutoInt's top-40 detected interactions in AutoInt | AutoInt · Criteo | logloss: 0.4434 -> 0.4426±{2e-4} | Improved |
| Encode AutoInt's top-10 detected interactions in AutoInt | AutoInt · Avazu | AUC: 0.7774 -> 0.7773±1e-4 | No clear effect |
| Encode AutoInt's top-10 detected interactions in AutoInt | AutoInt · Avazu | logloss: 0.3811 -> 0.3811±5e-5 | No clear effect |
| Add detected feature interactions up to rank threshold L | MADEX explanations · DNA-CNN test instances | mean-squared error: 10e-3±1e-3 -> 5.0e-3±8e-4 | Improved |
| Add detected feature interactions up to rank threshold L | MADEX explanations · Sentiment-LSTM test instances | mean-squared error: 8.0e-2±6e-3 -> 0.4e-2±3e-3 | Improved |
| Add detected feature interactions up to rank threshold L | MADEX explanations · ResNet152 test instances | mean-squared error: 1.9±0.1 -> 0.9±0.2 | Improved |
| Add detected node interactions up to rank threshold L | MADEX explanations · GCN Cora test instances | mean-squared error: 7.1e3±7e2 -> 2e3±1e3 | Improved |
Closest papers by meaning
- From Feature Interaction to Feature Generation: A Generative Paradigm of CTR Prediction Models (ICML 2025 · similarity 0.84)
- On the Embedding Collapse when Scaling up Recommendation Models (ICML 2024 · similarity 0.84)
- Language Representations Can be What Recommenders Need: Findings and Potentials (ICLR 2025 · similarity 0.84)
- BeFA: A General Behavior-driven Feature Adapter for Multimedia Recommendation (AAAI 2025 · similarity 0.84)
- Improving LLMs for Recommendation with Out-Of-Vocabulary Tokens (ICML 2025 · similarity 0.84)
- Coherency Improved Explainable Recommendation via Large Language Model (AAAI 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.
