Selected work

Representative Papers

Attention-Enhanced U-Net for Accurate Segmentation of COVID-19 Infected Lung Regions in CT Scans

May 18, 2025

To address insufficient segmentation accuracy of infected lung regions in COVID-19 CT images, this paper proposes an enhanced U-Net architecture incorporating a Channel-and-Spatial Coordinated Attention Mechanism (CBAM). The CBAM module is innovatively embedded into both encoder and decoder pathways to jointly strengthen perception of subtle lesion microstructures and suppress background interference. Additionally, multi-scale data augmentation and morphological post-processing are integrated to improve model robustness. Evaluated on a public benchmark dataset, the proposed method achieves a Dice coefficient of 0.8658 and a mean IoU of 0.8316—significantly outperforming standard U-Net and state-of-the-art baseline models. This work delivers a high-precision, interpretable segmentation tool for clinical decision support in COVID-19 diagnosis.

0 citationsRead paper

Towards Recommender Systems LLMs Playground (RecSysLLMsP): Exploring Polarization and Engagement in Simulated Social Networks

Jan 29, 2025

This study investigates the causal mechanisms by which recommendation strategies influence user engagement and opinion polarization, aiming to balance individual satisfaction with societal cognitive diversity. We propose RecSysLLMsP—a novel framework leveraging large language models (LLMs) to construct a configurable multi-agent simulation environment. It introduces a sandbox-style testing platform for recommender systems, uniquely integrating LLM-driven agent prompts, and establishes a modeling paradigm that couples static and dynamic agent attributes. For the first time, we quantitatively assess the causal effects of distinct recommendation strategies—similarity-based, diversity-promoting, and balanced—on polarization within a controlled, reproducible simulation. Experimental results show that similarity-based recommendations significantly increase engagement but intensify echo chambers; diversity-oriented strategies foster cross-ideological interaction yet suffer from behavioral instability. The framework enables pre-deployment strategy evaluation and faithful replication of polarization dynamics, providing a verifiable, methodology-grounded foundation for responsible recommender system design.

0 citationsRead paper

Towards New Benchmark for AI Alignment&Sentiment Analysis in Socially Important Issues: A Comparative Study of Human and LLMs in the Context of AGI

Jan 05, 2025

This study investigates the affective understanding and dynamic evolution mechanisms of large language models (LLMs) regarding societally critical topics such as artificial general intelligence (AGI), aiming to advance scientifically grounded affective evaluation in AI alignment. Method: Leveraging Likert-scale–based human–AI comparative experiments, we systematically assess affective tendencies and three-day temporal dynamics across seven mainstream LLMs (e.g., GPT-4, Bard) and three human cohorts. Contribution/Results: We first reveal significant heterogeneity in LLM affective distributions on AGI—alongside quantifiable temporal evolution (evolution rate differences: 1.03%–8.21%)—and find that LLMs’ mean affective scores (3.32–4.12/5) significantly exceed the human average (2.97/5), exposing latent biases and conflict-of-interest risks. We propose the “human-like but non-uniform” hypothesis for LLM affect formation and introduce the first AI affective alignment benchmark tailored to societally salient issues.

0 citationsRead paper

Recent publications

Latest Papers

Attention-Enhanced U-Net for Accurate Segmentation of COVID-19 Infected Lung Regions in CT Scans

May 18, 2025

To address insufficient segmentation accuracy of infected lung regions in COVID-19 CT images, this paper proposes an enhanced U-Net architecture incorporating a Channel-and-Spatial Coordinated Attention Mechanism (CBAM). The CBAM module is innovatively embedded into both encoder and decoder pathways to jointly strengthen perception of subtle lesion microstructures and suppress background interference. Additionally, multi-scale data augmentation and morphological post-processing are integrated to improve model robustness. Evaluated on a public benchmark dataset, the proposed method achieves a Dice coefficient of 0.8658 and a mean IoU of 0.8316—significantly outperforming standard U-Net and state-of-the-art baseline models. This work delivers a high-precision, interpretable segmentation tool for clinical decision support in COVID-19 diagnosis.

0 citationsRead paper

Towards Recommender Systems LLMs Playground (RecSysLLMsP): Exploring Polarization and Engagement in Simulated Social Networks

Jan 29, 2025

This study investigates the causal mechanisms by which recommendation strategies influence user engagement and opinion polarization, aiming to balance individual satisfaction with societal cognitive diversity. We propose RecSysLLMsP—a novel framework leveraging large language models (LLMs) to construct a configurable multi-agent simulation environment. It introduces a sandbox-style testing platform for recommender systems, uniquely integrating LLM-driven agent prompts, and establishes a modeling paradigm that couples static and dynamic agent attributes. For the first time, we quantitatively assess the causal effects of distinct recommendation strategies—similarity-based, diversity-promoting, and balanced—on polarization within a controlled, reproducible simulation. Experimental results show that similarity-based recommendations significantly increase engagement but intensify echo chambers; diversity-oriented strategies foster cross-ideological interaction yet suffer from behavioral instability. The framework enables pre-deployment strategy evaluation and faithful replication of polarization dynamics, providing a verifiable, methodology-grounded foundation for responsible recommender system design.

0 citationsRead paper

Towards New Benchmark for AI Alignment&Sentiment Analysis in Socially Important Issues: A Comparative Study of Human and LLMs in the Context of AGI

Jan 05, 2025

This study investigates the affective understanding and dynamic evolution mechanisms of large language models (LLMs) regarding societally critical topics such as artificial general intelligence (AGI), aiming to advance scientifically grounded affective evaluation in AI alignment. Method: Leveraging Likert-scale–based human–AI comparative experiments, we systematically assess affective tendencies and three-day temporal dynamics across seven mainstream LLMs (e.g., GPT-4, Bard) and three human cohorts. Contribution/Results: We first reveal significant heterogeneity in LLM affective distributions on AGI—alongside quantifiable temporal evolution (evolution rate differences: 1.03%–8.21%)—and find that LLMs’ mean affective scores (3.32–4.12/5) significantly exceed the human average (2.97/5), exposing latent biases and conflict-of-interest risks. We propose the “human-like but non-uniform” hypothesis for LLM affect formation and introduce the first AI affective alignment benchmark tailored to societally salient issues.

0 citationsRead paper