Get started
Work

Learning human aligned explanations through supervised aggregation of XAI methods for Arabic sentiment analysis

Year: 2026
Type: article
Abstract: Explainable Artificial Intelligence (XAI) aims to make complex AI models moretransparent by revealing how their predictions are made. However, XAI methods such as LIME and SHAP often yield inconsistent results, as each captures different aspects of model behavior. Previous ensemble approaches, including our earlier Ensemble XAI based on simple averaging of token-importance scores from different XAI methods, improved stability but failed to learn human preferred weightings or capture nonlinear relationships among methods. In this work, we propose a meta-model framework that learns to combine XAI outputs under human supervision for Arabic sentiment analysis. Using human annotated datasets and a transformer model as the base black-box model, we train three supervised meta-models (Logistic Regression, XGBoost and CatBoost) on token-level features derived from multiple XAI scores, transformer attention weights, embeddings, and other linguistic features. Results show that the meta-model substantially improves plausibility (human alignment), Catboost meta-model reaches an F1 score of 0.84 compared to 0.64-0.70 for individual XAI methods such as LIME, SHAP, IG, DeepLIFT and Ensemble XAI.This gain comes with a trade-off in faithfulness, where the meta-models scores are lower than our previously proposed Ensemble XAI on our combined faithfulness metric (0.61 for XGBboost meta-model vs. 0.76 for Ensemble XAI).This study provides the first investigation of human-supervised XAI aggregation in Arabic, offering new insights into balancing faithfulness and plausibility in explainable NLP. The code implementation can be publicly accessible at: https://github.com/ychafiqui/xai-meta-model
DOI: 10.1007/s44163-026-02193-0
Language: English
Open Access status: gold
Estimated APC paid: $1,590
OpenAlex ID: W7212133039
FWCI:
Cites: 14
Cited by:
Related to: