Selected work

Representative Papers

MLPlatt: Simple Calibration Framework for Ranking Models

Jan 13, 2026

This work addresses the challenge that e-commerce ranking models often produce poorly calibrated and uninterpretable outputs, hindering reliable click-through rate (CTR) probability estimation. To resolve this, the authors propose a context-aware post-hoc calibration method that transforms model scores into well-calibrated, interpretable probabilities while preserving the original ranking order. The approach further enables stratified conditional calibration across categorical attributes such as country or device type, achieving accurate probability estimates at both global and fine-grained levels without compromising ranking performance. Experimental results on two real-world datasets demonstrate that the proposed method reduces the feature-based Expected Calibration Error (F-ECE) by over 10% compared to existing techniques, effectively meeting the practical demand in e-commerce for trustworthy probabilistic predictions.

0 citationsRead paper

Divide (Text) and Conquer (Sentiment): Improved Sentiment Classification by Constituent Conflict Resolution

May 08, 2025

To address the degradation of sentiment classification performance on long texts caused by multiple conflicting sentiments, this paper proposes a constituent-level sentiment decomposition-aggregation paradigm. Methodologically, it first performs fine-grained sentiment constituent extraction and local conflict identification based on syntactic structure—without introducing additional parameters—and then employs a lightweight, learnable MLP aggregator to adaptively fuse conflicting sentiments. This framework is the first to explicitly model conflict resolution as a constituent-level decomposition task. It achieves significant improvements over state-of-the-art methods across multiple benchmarks—including Amazon, Twitter, and SST—yielding up to a 12.7% F1-score gain on long sentences containing conflicting expressions. Moreover, it improves training efficiency by 100× compared to full-model fine-tuning baselines and reduces computational cost to just 1% of that baseline.

0 citationsRead paper

Recent publications

Latest Papers

MLPlatt: Simple Calibration Framework for Ranking Models

Jan 13, 2026

This work addresses the challenge that e-commerce ranking models often produce poorly calibrated and uninterpretable outputs, hindering reliable click-through rate (CTR) probability estimation. To resolve this, the authors propose a context-aware post-hoc calibration method that transforms model scores into well-calibrated, interpretable probabilities while preserving the original ranking order. The approach further enables stratified conditional calibration across categorical attributes such as country or device type, achieving accurate probability estimates at both global and fine-grained levels without compromising ranking performance. Experimental results on two real-world datasets demonstrate that the proposed method reduces the feature-based Expected Calibration Error (F-ECE) by over 10% compared to existing techniques, effectively meeting the practical demand in e-commerce for trustworthy probabilistic predictions.

0 citationsRead paper

Divide (Text) and Conquer (Sentiment): Improved Sentiment Classification by Constituent Conflict Resolution

May 08, 2025

To address the degradation of sentiment classification performance on long texts caused by multiple conflicting sentiments, this paper proposes a constituent-level sentiment decomposition-aggregation paradigm. Methodologically, it first performs fine-grained sentiment constituent extraction and local conflict identification based on syntactic structure—without introducing additional parameters—and then employs a lightweight, learnable MLP aggregator to adaptively fuse conflicting sentiments. This framework is the first to explicitly model conflict resolution as a constituent-level decomposition task. It achieves significant improvements over state-of-the-art methods across multiple benchmarks—including Amazon, Twitter, and SST—yielding up to a 12.7% F1-score gain on long sentences containing conflicting expressions. Moreover, it improves training efficiency by 100× compared to full-model fine-tuning baselines and reduces computational cost to just 1% of that baseline.

0 citationsRead paper