arXiv subjects

Huimin Xu

Publications and source records attributed to Huimin Xu.

At least 19 recordsLinked to original sources

arXiv2026

Understanding and Preventing Entropy Collapse in RLVR with On-Policy Entropy Flow Optimization

Reinforcement learning with verifiable rewards (RLVR) has become an effective paradigm for improving the reasoning ability of large language models. However, widely used RLVR algorithms, such as GRPO, often suffer from entropy collapse, leading to premature determinism and unstable optimization. Existing remedies, including entropy regularization and ratio-based clipping heuristics, either control entropy in a coarse-grained manner or rely on approximate on-policy training. In this paper, we revisit entropy collapse from a token-level entropy flow perspective. Our analysis reveals that entropy-decreasing tokens consistently outweigh entropy-increasing ones, resulting in a severely imbalanced entropy flow. This perspective provides a unified explanation of entropy collapse in existing RLVR algorithms and highlights the importance of balancing entropy dynamics. Motivated by this analysis, we propose On-Policy Entropy Flow Optimization (OPEFO), an adaptive entropy flow balancing mechanism that rescales entropy-increasing and entropy-decreasing updates according to their contributions to entropy change, while remaining strict on-policy. Experiments on six mathematical reasoning benchmarks demonstrate that OPEFO improves training stability and final performance. We will release the code and models upon publication.

cs.LG↗

arXiv2026

Automated Thematic Analysis for Clinical Qualitative Data: Iterative Codebook Refinement with Full Provenance

Thematic analysis (TA) is widely used in health research to extract patterns from patient interviews, yet manual TA faces challenges in scalability and reproducibility. LLM-based automation can help, but existing approaches produce codebooks with limited generalizability and lack analytic auditability. We present an automated TA framework combining iterative codebook refinement with full provenance tracking. Evaluated on five corpora spanning clinical interviews, social media, and public transcripts, the framework achieves the highest composite quality score on four of five datasets compared to six baselines. Iterative refinement yields statistically significant improvements on four datasets with large effect sizes, driven by gains in code reusability and distributional consistency while preserving descriptive quality. On two clinical corpora (pediatric cardiology), generated themes align with expert-annotated themes.

cs.CL↗

arXiv2025

SFT-TA: Supervised Fine-Tuned Agents in Multi-Agent LLMs for Automated Inductive Thematic Analysis

Thematic Analysis (TA) is a widely used qualitative method that provides a structured yet flexible framework for identifying and reporting patterns in clinical interview transcripts. However, manual thematic analysis is time-consuming and limits scalability. Recent advances in LLMs offer a pathway to automate thematic analysis, but alignment with human results remains limited. To address these limitations, we propose SFT-TA, an automated thematic analysis framework that embeds supervised fine-tuned (SFT) agents within a multi-agent system. Our framework outperforms existing frameworks and the gpt-4o baseline in alignment with human reference themes. We observed that SFT agents alone may underperform, but achieve better results than the baseline when embedded within a multi-agent system. Our results highlight that embedding SFT agents in specific roles within a multi-agent system is a promising pathway to improve alignment with desired outputs for thematic analysis.

cs.CL↗

arXiv2025

Auto-TA: Towards Scalable Automated Thematic Analysis (TA) via Multi-Agent Large Language Models with Reinforcement Learning

Congenital heart disease (CHD) presents complex, lifelong challenges often underrepresented in traditional clinical metrics. While unstructured narratives offer rich insights into patient and caregiver experiences, manual thematic analysis (TA) remains labor-intensive and unscalable. We propose a fully automated large language model (LLM) pipeline that performs end-to-end TA on clinical narratives, which eliminates the need for manual coding or full transcript review. Our system employs a novel multi-agent framework, where specialized LLM agents assume roles to enhance theme quality and alignment with human analysis. To further improve thematic relevance, we optionally integrate reinforcement learning from human feedback (RLHF). This supports scalable, patient-centered analysis of large qualitative datasets and allows LLMs to be fine-tuned for specific clinical contexts.

cs.CL↗

arXiv2025

SCOPE: Compress Mathematical Reasoning Steps for Efficient Automated Process Annotation

Process Reward Models (PRMs) have demonstrated promising results in mathematical reasoning, but existing process annotation approaches, whether through human annotations or Monte Carlo simulations, remain computationally expensive. In this paper, we introduce Step COmpression for Process Estimation (SCOPE), a novel compression-based approach that significantly reduces annotation costs. We first translate natural language reasoning steps into code and normalize them through Abstract Syntax Tree, then merge equivalent steps to construct a prefix tree. Unlike simulation-based methods that waste numerous samples on estimation, SCOPE leverages a compression-based prefix tree where each root-to-leaf path serves as a training sample, reducing the complexity from $O(NMK)$ to $O(N)$. We construct a large-scale dataset containing 196K samples with only 5% of the computational resources required by previous methods. Empirical results demonstrate that PRMs trained on our dataset consistently outperform existing automated annotation approaches on both Best-of-N strategy and ProcessBench.

cs.AI↗

arXiv2025

Adapting to LLMs: How Insiders and Outsiders Reshape Scientific Knowledge Production

CSCW has long examined how emerging technologies reshape the ways researchers collaborate and produce knowledge, with scientific knowledge production as a central area of focus. As AI becomes increasingly integrated into scientific research, understanding how researchers adapt to it reveals timely opportunities for CSCW research -- particularly in supporting new forms of collaboration, knowledge practices, and infrastructure in AI-driven science. This study quantifies LLM impacts on scientific knowledge production based on an evaluation workflow that combines an insider-outsider perspective with a knowledge production framework. Our findings reveal how LLMs catalyze both innovation and reorganization in scientific communities, offering insights into the broader transformation of knowledge production in the age of generative AI and sheds light on new research opportunities in CSCW.

cs.HC↗

arXiv2025

TAMA: A Human-AI Collaborative Thematic Analysis Framework Using Multi-Agent LLMs for Clinical Interviews

Thematic analysis (TA) is a widely used qualitative approach for uncovering latent meanings in unstructured text data. TA provides valuable insights in healthcare but is resource-intensive. Large Language Models (LLMs) have been introduced to perform TA, yet their applications in high-stakes healthcare settings, particularly for qualitative clinical interview analysis, remain limited. Here, we propose TAMA: A Human-AI Collaborative Thematic Analysis framework using Multi-Agent LLMs for clinical interviews. We leverage the scalability and coherence of multi-agent systems through structured conversations between agents and coordinate the expertise of cardiac experts in TA. Using interview transcripts from parents of children with Anomalous Aortic Origin of a Coronary Artery (AAOCA), a rare congenital heart disease, we demonstrate that TAMA outperforms single-agent LLM TA approaches, achieving higher thematic hit rate, coverage, and distinctiveness. TAMA demonstrates strong potential for automated TA in clinical settings by leveraging multi-agent LLM systems with human-in-the-loop integration by enhancing quality while significantly reducing manual workload. The full implementation is publicly available at https://github.com/Charlie-Yi-SJ/TAMA.

cs.HC↗

arXiv2025

Full-Step-DPO: Self-Supervised Preference Optimization with Step-wise Rewards for Mathematical Reasoning

Direct Preference Optimization (DPO) often struggles with long-chain mathematical reasoning. Existing approaches, such as Step-DPO, typically improve this by focusing on the first erroneous step in the reasoning chain. However, they overlook all other steps and rely heavily on humans or GPT-4 to identify erroneous steps. To address these issues, we propose Full-Step-DPO, a novel DPO framework tailored for mathematical reasoning. Instead of optimizing only the first erroneous step, it leverages step-wise rewards from the entire reasoning chain. This is achieved by training a self-supervised process reward model, which automatically scores each step, providing rewards while avoiding reliance on external signals. Furthermore, we introduce a novel step-wise DPO loss, which dynamically updates gradients based on these step-wise rewards. This endows stronger reasoning capabilities to language models. Extensive evaluations on both in-domain and out-of-domain mathematical reasoning benchmarks across various base language models, demonstrate that Full-Step-DPO achieves superior performance compared to state-of-the-art baselines.

cs.CL↗

arXiv2024

As Simple as Fine-tuning: LLM Alignment via Bidirectional Negative Feedback Loss

Direct Preference Optimization (DPO) has emerged as a more computationally efficient alternative to Reinforcement Learning from Human Feedback (RLHF) with Proximal Policy Optimization (PPO), eliminating the need for reward models and online sampling. Despite these benefits, DPO and its variants remain sensitive to hyper-parameters and prone to instability, particularly on mathematical datasets. We argue that these issues arise from the unidirectional likelihood-derivative negative feedback inherent in the log-likelihood loss function. To address this, we propose a novel LLM alignment loss that establishes a stable Bidirectional Negative Feedback (BNF) during optimization. Our proposed BNF loss eliminates the need for pairwise contrastive losses and does not require any extra tunable hyper-parameters or pairwise preference data, streamlining the alignment pipeline to be as simple as supervised fine-tuning. We conduct extensive experiments across two challenging QA benchmarks and four reasoning benchmarks. The experimental results show that BNF achieves comparable performance to the best methods on QA benchmarks, while its performance decrease on the four reasoning benchmarks is significantly lower compared to the best methods, thus striking a better balance between value alignment and reasoning ability. In addition, we further validate the performance of BNF on non-pairwise datasets, and conduct in-depth analysis of log-likelihood and logit shifts across different preference optimization methods.

cs.CL↗

arXiv2024

How high-status women promote repeated collaboration among women in male-dominated contexts

Male-dominated contexts pose a dilemma: they increase the benefits of repeated collaboration among women, yet at the same time, make such collaborations less likely. This paper seeks to understand the conditions that foster repeated collaboration among women versus men in male-dominated settings by examining the critical role of status hierarchies. Using collaboration data on 8,232,769 computer science research teams, we found that when a woman holds the top-ranking position in a steep status hierarchy, other women on that team are more likely than men to collaborate again, as compared to when the hierarchy is flat, and compared to when men occupy the top-ranking position. In steep hierarchies, top-ranking women but not top-ranking men foster conditions in which junior women are more likely to collaborate again than junior men of similar status levels. Our research suggests that whereas status hierarchies are especially detrimental to repeated collaboration among underrepresented individuals, top-ranking women in steep status hierarchies mitigate these negative impacts between women in male-dominated settings.

cs.CY↗

arXiv2024

Negation Triplet Extraction with Syntactic Dependency and Semantic Consistency

Previous works of negation understanding mainly focus on negation cue detection and scope resolution, without identifying negation subject which is also significant to the downstream tasks. In this paper, we propose a new negation triplet extraction (NTE) task which aims to extract negation subject along with negation cue and scope. To achieve NTE, we devise a novel Syntax&Semantic-Enhanced Negation Extraction model, namely SSENE, which is built based on a generative pretrained language model (PLM) {of Encoder-Decoder architecture} with a multi-task learning framework. Specifically, the given sentence's syntactic dependency tree is incorporated into the PLM's encoder to discover the correlations between the negation subject, cue and scope. Moreover, the semantic consistency between the sentence and the extracted triplet is ensured by an auxiliary task learning. Furthermore, we have constructed a high-quality Chinese dataset NegComment based on the users' reviews from the real-world platform of Meituan, upon which our evaluations show that SSENE achieves the best NTE performance compared to the baselines. Our ablation and case studies also demonstrate that incorporating the syntactic information helps the PLM's recognize the distant dependency between the subject and cue, and the auxiliary task learning is helpful to extract the negation triplets with more semantic consistency.

cs.CL↗

arXiv2024

Don't Forget Your Reward Values: Language Model Alignment via Value-based Calibration

While Reinforcement Learning from Human Feedback (RLHF) significantly enhances the generation quality of Large Language Models (LLMs), recent studies have raised concerns regarding the complexity and instability associated with the Proximal Policy Optimization (PPO) algorithm, proposing a series of order-based calibration methods as viable alternatives. This paper delves further into current order-based methods, examining their inefficiencies in utilizing reward values and addressing misalignment issues. Building upon these findings, we propose a novel \textbf{V}alue-based \textbf{C}ali\textbf{B}ration (VCB) method to better align LLMs with human preferences. Experimental results demonstrate that VCB surpasses existing alignment methods on AI assistant and summarization datasets, providing impressive generalizability, robustness, and stability in diverse settings.

cs.CL↗

arXiv2024

Exploiting Duality in Open Information Extraction with Predicate Prompt

Open information extraction (OpenIE) aims to extract the schema-free triplets in the form of (\emph{subject}, \emph{predicate}, \emph{object}) from a given sentence. Compared with general information extraction (IE), OpenIE poses more challenges for the IE models, {especially when multiple complicated triplets exist in a sentence. To extract these complicated triplets more effectively, in this paper we propose a novel generative OpenIE model, namely \emph{DualOIE}, which achieves a dual task at the same time as extracting some triplets from the sentence, i.e., converting the triplets into the sentence.} Such dual task encourages the model to correctly recognize the structure of the given sentence and thus is helpful to extract all potential triplets from the sentence. Specifically, DualOIE extracts the triplets in two steps: 1) first extracting a sequence of all potential predicates, 2) then using the predicate sequence as a prompt to induce the generation of triplets. Our experiments on two benchmarks and our dataset constructed from Meituan demonstrate that DualOIE achieves the best performance among the state-of-the-art baselines. Furthermore, the online A/B test on Meituan platform shows that 0.93\% improvement of QV-CTR and 0.56\% improvement of UV-CTR have been obtained when the triplets extracted by DualOIE were leveraged in Meituan's search system.

cs.CL↗

arXiv2023

The Impact of Heterogeneous Shared Leadership in Scientific Teams

Leadership is evolving dynamically from an individual endeavor to shared efforts. This paper aims to advance our understanding of shared leadership in scientific teams. We define three kinds of leaders, junior (10-15), mid (15-20), and senior (20+) based on career age. By considering the combinations of any two leaders, we distinguish shared leadership as heterogeneous when leaders are in different age cohorts and homogeneous when leaders are in the same age cohort. Drawing on 1,845,351 CS, 254,039 Sociology, and 193,338 Business teams with two leaders in the OpenAlex dataset, we identify that heterogeneous shared leadership brings higher citation impact for teams than homogeneous shared leadership. Specifically, when junior leaders are paired with senior leaders, it significantly increases team citation ranking by 1-2%, in comparison with two leaders of similar age. We explore the patterns between homogeneous leaders and heterogeneous leaders from team scale, expertise composition, and knowledge recency perspectives. Compared with homogeneous leaders, heterogeneous leaders are more adaptive in large teams, have more diverse expertise, and trace both the newest and oldest references.

physics.soc-ph↗

arXiv2023

Disparity in the Evolving COVID-19 Collaboration Network

The COVID 19 pandemic has paused many ongoing research projects and unified researchers' attention to focus on COVID 19 related issues. Our project traces 712294 scientists' publications related to COVID 19 for two years, from January 2020 to December 2021, to detect the dynamic evolution patterns of the COVID 19 collaboration network over time. By studying the collaboration network of COVID 19 scientists, we observe how a new scientific community has been built in preparation for a sudden shock. The number of newcomers grows incrementally, and the connectivity of the collaboration network shifts from loose to tight promptly. Even though every scientist has an equal opportunity to start a study, collaboration disparity still exists. Following the scale-free distribution, only a few top authors are highly connected with other authors. These top authors are more likely to attract newcomers and work with each other. As the collaboration network evolves, the increase rate in the probability of attracting newcomers for authors with higher degrees increases, whereas the increase rates in the likelihood of forming new links among authors with higher degrees decreases. This highlights the interesting trend that the COVID pandemic alters the research collaboration trends that star scientists are starting to collaborate more with newcomers but less with existing collaborators, which, in a certain way, reduces the collaboration disparity.

cs.SI↗

arXiv2022

Team Power Dynamics and Team Impact: New Perspectives on Scientific Collaboration using Career Age as a Proxy for Team Power

Power dynamics influence every aspect of scientific collaboration. Team power dynamics can be measured by team power level and team power hierarchy. Team power level is conceptualized as the average level of the possession of resources, expertise, or decision-making authorities of a team. Team power hierarchy represents the vertical differences of the possessions of resources in a team. In Science of Science, few studies have looked at scientific collaboration from the perspective of team power dynamics. This research examines how team power dynamics affect team impact to fill the research gap. In this research, all co-authors of one publication are treated as one team. Team power level and team power hierarchy of one team are measured by the mean and Gini index of career age of co-authors in this team. Team impact is quantified by citations of a paper authored by this team. By analyzing over 7.7 million teams from Science (e.g., Computer Science, Physics), Social Sciences (e.g., Sociology, Library & Information Science), and Arts & Humanities (e.g., Art), we find that flat team structure is associated with higher team impact, especially when teams have high team power level. These findings have been repeated in all five disciplines except Art, and are consistent in various types of teams from Computer Science including teams from industry or academia, teams with different gender groups, teams with geographical contrast, and teams with distinct size.

physics.soc-ph↗

arXiv2020

The Geometry of Information Cocoon: Analyzing the Cultural Space with Word Embedding Models

Accompanied by the development of digital media, the threat of information cocoon has become a significant issue. However, little is known about the measure of information cocoon as a cultural space and its relationship with social class. This study addresses this problem by constructing the cultural space with word embedding models and random shuffling methods among three large-scale digital media use datasets. In the light of field theory of cultural production, we investigate the information cocoon effect on different social classes among 979 computer users, 100,000 smartphone users, and 159,373 mobile reading application users. Our analysis reveals that information cocoons widely exist in the daily use of digital media. Moreover, people of lower social class have a higher probability of getting stuck in the information cocoon filled with the entertainment content. In contrast, the people of higher social class have more capability to stride over the constraints of the information cocoon. The results suggest that the disadvantages for vulnerable groups in acquiring knowledge may further widen social inequality.

cs.CY↗

arXiv2020

Relational Reflection Entity Alignment

Entity alignment aims to identify equivalent entity pairs from different Knowledge Graphs (KGs), which is essential in integrating multi-source KGs. Recently, with the introduction of GNNs into entity alignment, the architectures of recent models have become more and more complicated. We even find two counter-intuitive phenomena within these methods: (1) The standard linear transformation in GNNs is not working well. (2) Many advanced KG embedding models designed for link prediction task perform poorly in entity alignment. In this paper, we abstract existing entity alignment methods into a unified framework, Shape-Builder & Alignment, which not only successfully explains the above phenomena but also derives two key criteria for an ideal transformation operation. Furthermore, we propose a novel GNNs-based method, Relational Reflection Entity Alignment (RREA). RREA leverages Relational Reflection Transformation to obtain relation specific embeddings for each entity in a more efficient way. The experimental results on real-world datasets show that our model significantly outperforms the state-of-the-art methods, exceeding by 5.8%-10.9% on Hits@1.

cs.IR↗