Research map / Multi-agent systems and games / Human-AI and goal-conditioned
Counterfactual Explanations of Time Varying Rankings (Student Abstract)
Read the paperCheck your idea against 108,967 papers
AI-extracted summary. Written by AI from the paper's full text. Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.
Proposes CFETVR, a counterfactual explanation method for rankings that forecasts all agents' future features with ARIMA, re-ranks with XGBoost, and greedily searches feature changes for a target agent.
Problem
Existing counterfactual explanations for rankings adjust only one agent's features and ignore how other agents' rankings shift over time, and per-agent feature searches over all agents can cause combinatorial explosion.
Approach
Re-evaluate the ranking using ARIMA forecasts of all agents' future features, then run a greedy feature search (guided by SHAP important features and interactions) on the target agent so the counterfactual accounts for the dynamic ranking system.
Key results
- On the Walmart dataset with 45 stores, the target store (ID 23) moved from rank 10 to rank 5 after the proposed counterfactual search, with a required store size of 87,972 sq ft.
- In some low-change scenarios, the ARIMA forecast alone moved store ID 7 up two places, meeting the target without any feature search.
Measured effects reported in the paper
| What was changed | Where | Measured | Effect |
|---|---|---|---|
| Forecast all agents’ future features with ARIMA and search for target-agent feature changes | CFETVR with XGBoost · Walmart sales prediction dataset, store ID 23 | rank: 10 -> 5 | Helped |
| Forecast other agents’ features with ARIMA before feature search | CFETVR · Walmart sales prediction dataset, store ID 7 | rank: 43rd by one position; increase the rank by two places | Helped |
Surprising findings
- Changes in predicted rankings were observed across all scenarios but each agent's rank shifted only modestly.
Limitations
- Assumes feature changes are independent; feature dependencies are not modeled.
- Evaluation is a small set of illustrative experiments on one dataset with no quantitative baselines.
Experimental setup
Datasets: Walmart sales prediction dataset (Kaggle)
Benchmarks: Walmart sales prediction dataset
Metrics: rank change achieved counterfactual feature change
Baselines: Salimiparsa 2023 minimum-feature-change ranking counterfactual
Builds on: ARIMA SHAP XGBoost
Closest papers by meaning
- Generating Likely Counterfactuals Using Sum-Product Networks (ICLR 2025 · similarity 0.87)
- Generating Counterfactual Explanations Under Temporal Constraints (AAAI 2025 · similarity 0.87)
- Understanding Individual Agent Importance in Multi-Agent System via Counterfactual Reasoning (AAAI 2025 · similarity 0.86)
- Semi-factual Explanations in AI (AAAI 2024 · similarity 0.85)
- Even-if Explanations: Formal Foundations, Priorities and Complexity (AAAI 2025 · similarity 0.85)
- Counterfactual Explanations for Misclassified Images: How Human and Machine Explanations Differ (Abstract Reprint) (AAAI 2024 · similarity 0.85)
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.