What Makes a 3D Scene Editable? A Factorized Benchmark of Fidelity, Locality, Consistency, and Preservation
研究通过引入EditBench3D评估3D场景编辑的四个属性：保真度、局部性、一致性和非目标内容保留，使用多种编辑方法并揭示了各方法在不同维度上的表现差异。
0 citationsRead paper
Selected work
Representative Papers
Reexamining zero-shot summarization: Empirical investigation of trustworthiness of LLM-summarizers
This study addresses the output stochasticity of large language models (LLMs) in zero-shot summarization, which undermines result stability and compromises trustworthiness—particularly in high-stakes domains such as education. To systematically evaluate the reliability of LLM-based summarizers, the authors propose a two-tier diagnostic protocol comprising document-level stability analysis and a corpus-wide stability index. The approach innovatively integrates multidimensional stability coefficients with a stratified-sampling-based stability index, forming a quantifiable empirical framework that leverages controlled repeated generation, semantic and factual consistency scoring, and cross-genre evaluation. Experimental results across three document types reveal significant variability in summarization stability among different LLMs, demonstrating that stability is a critical determinant of model credibility.
0 citationsRead paper
PRecG: Legal Precedent Retrieval with Graph Neural Networks and Rhetorical Role Segmentation
This work addresses the limitations of existing legal case retrieval methods, which treat judicial documents as monolithic texts and overlook their rhetorical structure, thereby failing to capture nuanced semantic differences of legal entities across contexts. To overcome this, the authors propose a hierarchical modeling approach that first segments judgments into semantic units based on rhetorical roles, then constructs knowledge graphs for each segment and employs graph neural networks to learn context-aware representations of legal entities. These representations are hierarchically aggregated to produce paragraph- and document-level embeddings for computing semantic similarity between cases. By integrating rhetorical structure analysis with graph neural networks, this method enables fine-grained modeling of contextual semantics of legal concepts. Experiments on an Indian legal benchmark dataset demonstrate significant performance gains over state-of-the-art approaches, substantially improving case retrieval accuracy.
0 citationsRead paper
Financial Frequency Combs
This study investigates whether a fractional-order financial system endowed with long-term memory exhibits a discrete, equally spaced spectral structure—reminiscent of physical frequency combs—in its steady state. Building upon the Huang–Li–Ma–Chen fractional-order model formulated with Caputo derivatives, the authors combine spectral analysis and parameter sensitivity studies to demonstrate, for the first time, that such a stable frequency comb emerges within specific parameter regimes when the fractional order exceeds a critical threshold. This spectral structure proves robust against perturbations in initial interest rates and investment levels, though it is sensitive to the initial price level, and transitions to chaos as the fractional order increases further. These findings suggest that long-term economic cycles may manifest as deterministic discrete spectra rather than the traditionally assumed stochastic continuous spectra, offering a novel perspective on macroeconomic memory effects.
0 citationsRead paper
Beyond Metadata: The Role of DOI, ORCID, and ROR in Shaping Transparent and Interoperable Library Systems
This study addresses the limitations of library systems in digital scholarly communication, where metadata silos, insufficient interoperability, and weak adoption of persistent identifiers (PIDs) hinder effective linkage and sharing of research outputs. For the first time, the project conceptualizes DOIs, ORCID iDs, and ROR IDs as “connective infrastructures,” employing metadata analysis, global case comparisons, and PID literacy assessments to demonstrate their paradigm shift from static labels to machine-actionable, interconnected infrastructure. Findings reveal ORCID adoption rates of 41%–89% among German institutions, with poor metadata quality and low awareness identified as key barriers. Building on these insights, the study proposes best practices and innovative pathways for libraries, universities, and policymakers to collaboratively develop a robust PID ecosystem across regions, particularly in Europe and Latin America.
0 citationsRead paper
Recent publications
Latest Papers
What Makes a 3D Scene Editable? A Factorized Benchmark of Fidelity, Locality, Consistency, and Preservation
研究通过引入EditBench3D评估3D场景编辑的四个属性：保真度、局部性、一致性和非目标内容保留，使用多种编辑方法并揭示了各方法在不同维度上的表现差异。
0 citationsRead paper
Reexamining zero-shot summarization: Empirical investigation of trustworthiness of LLM-summarizers
This study addresses the output stochasticity of large language models (LLMs) in zero-shot summarization, which undermines result stability and compromises trustworthiness—particularly in high-stakes domains such as education. To systematically evaluate the reliability of LLM-based summarizers, the authors propose a two-tier diagnostic protocol comprising document-level stability analysis and a corpus-wide stability index. The approach innovatively integrates multidimensional stability coefficients with a stratified-sampling-based stability index, forming a quantifiable empirical framework that leverages controlled repeated generation, semantic and factual consistency scoring, and cross-genre evaluation. Experimental results across three document types reveal significant variability in summarization stability among different LLMs, demonstrating that stability is a critical determinant of model credibility.
0 citationsRead paper
PRecG: Legal Precedent Retrieval with Graph Neural Networks and Rhetorical Role Segmentation
This work addresses the limitations of existing legal case retrieval methods, which treat judicial documents as monolithic texts and overlook their rhetorical structure, thereby failing to capture nuanced semantic differences of legal entities across contexts. To overcome this, the authors propose a hierarchical modeling approach that first segments judgments into semantic units based on rhetorical roles, then constructs knowledge graphs for each segment and employs graph neural networks to learn context-aware representations of legal entities. These representations are hierarchically aggregated to produce paragraph- and document-level embeddings for computing semantic similarity between cases. By integrating rhetorical structure analysis with graph neural networks, this method enables fine-grained modeling of contextual semantics of legal concepts. Experiments on an Indian legal benchmark dataset demonstrate significant performance gains over state-of-the-art approaches, substantially improving case retrieval accuracy.
0 citationsRead paper
Financial Frequency Combs
This study investigates whether a fractional-order financial system endowed with long-term memory exhibits a discrete, equally spaced spectral structure—reminiscent of physical frequency combs—in its steady state. Building upon the Huang–Li–Ma–Chen fractional-order model formulated with Caputo derivatives, the authors combine spectral analysis and parameter sensitivity studies to demonstrate, for the first time, that such a stable frequency comb emerges within specific parameter regimes when the fractional order exceeds a critical threshold. This spectral structure proves robust against perturbations in initial interest rates and investment levels, though it is sensitive to the initial price level, and transitions to chaos as the fractional order increases further. These findings suggest that long-term economic cycles may manifest as deterministic discrete spectra rather than the traditionally assumed stochastic continuous spectra, offering a novel perspective on macroeconomic memory effects.
0 citationsRead paper
Beyond Metadata: The Role of DOI, ORCID, and ROR in Shaping Transparent and Interoperable Library Systems
This study addresses the limitations of library systems in digital scholarly communication, where metadata silos, insufficient interoperability, and weak adoption of persistent identifiers (PIDs) hinder effective linkage and sharing of research outputs. For the first time, the project conceptualizes DOIs, ORCID iDs, and ROR IDs as “connective infrastructures,” employing metadata analysis, global case comparisons, and PID literacy assessments to demonstrate their paradigm shift from static labels to machine-actionable, interconnected infrastructure. Findings reveal ORCID adoption rates of 41%–89% among German institutions, with poor metadata quality and low awareness identified as key barriers. Building on these insights, the study proposes best practices and innovative pathways for libraries, universities, and policymakers to collaboratively develop a robust PID ecosystem across regions, particularly in Europe and Latin America.
0 citationsRead paper