Single-Token Expected-Value Scoring for Cold-Start Candidate Ranking
为解决冷启动候选排序问题，提出单令牌期望值评分方法，通过微调小型语言模型并结合混合序数回归损失函数来提高求职者和雇主的相关性。
0 citationsRead paper
Selected work
Representative Papers
VERDI: Single-Call Confidence Estimation for Verification-Based LLM Judges via Decomposed Inference
This work addresses the unreliability of confidence estimation in current large language model (LLM) judge systems, which stems from the absence of token-level log probabilities and structured outputs. The authors propose a single-call confidence estimation method that requires no additional inference passes, leveraging decomposed chain-of-verification reasoning traces to extract three structured signals: Step-Verdict Alignment, Claim-Level Margin, and Evidence Grounding Score. The approach integrates decomposed reasoning, Platt-scaled logistic regression, and a lightweight natural language inference (NLI) model—used in place of regular expressions for evidence extraction. Experiments demonstrate substantial improvements over conventional logprob-based baselines, achieving AUROC scores of 0.66–0.91 on GPT-family models, elevating Qwen-family performance from poorly calibrated ranges of 0.32–0.49 to 0.56–0.70, and reaching 0.73–0.88 in production systems, with strong cross-model transferability.
0 citationsRead paper
Bias and Uncertainty in LLM-as-a-Judge Estimation
This study addresses systematic scoring biases in LLM-as-a-Judge evaluations, where existing correction methods suffer from cross-model calibration instability, often leading to directional misjudgments in model comparisons. Through theoretical analysis, Monte Carlo simulations, and experiments on MMLU-Pro, the work uncovers the hidden risks of shared calibration strategies and proposes two diagnostic metrics—judge quality (J) and cross-model calibration instability (ΔJ)—to assess the reliability of bias correction outcomes. The research successfully reproduces the sign-flipping phenomenon observed in prior work, empirically validating the effectiveness of the proposed indicators. Building on these insights, the authors establish a reporting protocol for LLM-as-a-Judge evaluations, offering both theoretical grounding and practical guidance to enhance the trustworthiness of comparative model assessments.
0 citationsRead paper
Kd-tree Based Wasserstein Distance Approximation for High-Dimensional Data
This work addresses the challenge of scaling Wasserstein distance computation to large-scale retrieval in high dimensions, where the cubic time complexity of exact computation is prohibitive and existing tree-based approximations—such as quadtrees—suffer from limited depth, high preprocessing overhead, and insufficient accuracy in high-dimensional settings. To overcome these limitations, the authors propose kd-Flowtree, the first method to integrate kd-trees into Wasserstein distance approximation by constructing optimal transport on a kd-tree embedding. This approach significantly reduces preprocessing time and mitigates the curse of dimensionality due to sparsity. Theoretical analysis provides a data-size-independent probabilistic upper bound on nearest neighbor search accuracy. Empirical results on real high-dimensional datasets demonstrate that kd-Flowtree simultaneously outperforms existing methods in both efficiency and approximation accuracy.
0 citationsRead paper
Recent publications
Latest Papers
Single-Token Expected-Value Scoring for Cold-Start Candidate Ranking
为解决冷启动候选排序问题，提出单令牌期望值评分方法，通过微调小型语言模型并结合混合序数回归损失函数来提高求职者和雇主的相关性。
0 citationsRead paper
VERDI: Single-Call Confidence Estimation for Verification-Based LLM Judges via Decomposed Inference
This work addresses the unreliability of confidence estimation in current large language model (LLM) judge systems, which stems from the absence of token-level log probabilities and structured outputs. The authors propose a single-call confidence estimation method that requires no additional inference passes, leveraging decomposed chain-of-verification reasoning traces to extract three structured signals: Step-Verdict Alignment, Claim-Level Margin, and Evidence Grounding Score. The approach integrates decomposed reasoning, Platt-scaled logistic regression, and a lightweight natural language inference (NLI) model—used in place of regular expressions for evidence extraction. Experiments demonstrate substantial improvements over conventional logprob-based baselines, achieving AUROC scores of 0.66–0.91 on GPT-family models, elevating Qwen-family performance from poorly calibrated ranges of 0.32–0.49 to 0.56–0.70, and reaching 0.73–0.88 in production systems, with strong cross-model transferability.
0 citationsRead paper
Bias and Uncertainty in LLM-as-a-Judge Estimation
This study addresses systematic scoring biases in LLM-as-a-Judge evaluations, where existing correction methods suffer from cross-model calibration instability, often leading to directional misjudgments in model comparisons. Through theoretical analysis, Monte Carlo simulations, and experiments on MMLU-Pro, the work uncovers the hidden risks of shared calibration strategies and proposes two diagnostic metrics—judge quality (J) and cross-model calibration instability (ΔJ)—to assess the reliability of bias correction outcomes. The research successfully reproduces the sign-flipping phenomenon observed in prior work, empirically validating the effectiveness of the proposed indicators. Building on these insights, the authors establish a reporting protocol for LLM-as-a-Judge evaluations, offering both theoretical grounding and practical guidance to enhance the trustworthiness of comparative model assessments.
0 citationsRead paper
Kd-tree Based Wasserstein Distance Approximation for High-Dimensional Data
This work addresses the challenge of scaling Wasserstein distance computation to large-scale retrieval in high dimensions, where the cubic time complexity of exact computation is prohibitive and existing tree-based approximations—such as quadtrees—suffer from limited depth, high preprocessing overhead, and insufficient accuracy in high-dimensional settings. To overcome these limitations, the authors propose kd-Flowtree, the first method to integrate kd-trees into Wasserstein distance approximation by constructing optimal transport on a kd-tree embedding. This approach significantly reduces preprocessing time and mitigates the curse of dimensionality due to sparsity. Theoretical analysis provides a data-size-independent probabilistic upper bound on nearest neighbor search accuracy. Empirical results on real high-dimensional datasets demonstrate that kd-Flowtree simultaneously outperforms existing methods in both efficiency and approximation accuracy.
0 citationsRead paper