Selected work

Representative Papers

Enhancing Distributed Authorization With Lagrange Interpolation And Attribute-Based Encryption

Dec 25, 2025

To address the high authorization overhead and key distribution bottlenecks in ciphertext data access within distributed environments, this paper proposes a lightweight access control scheme integrating attribute-based encryption (ABE) with an optimized key sharing mechanism. Methodologically, it innovatively incorporates second-order Lagrange interpolation into Shamir’s secret sharing to enable low-overhead key reconstruction, and synergistically combines involution-based stream ciphers with ABE to achieve both efficient decryption and fine-grained policy-based authorization. The scheme is formally proven secure against collusion attacks and provides forward secrecy. Experimental evaluation demonstrates a 37% reduction in end-to-end encryption/decryption latency, a 42% decrease in server-side computational overhead, and a 2.1× improvement in system throughput compared to state-of-the-art approaches.

0 citationsRead paper

StrucSum: Graph-Structured Reasoning for Long Document Extractive Summarization with LLMs

May 29, 2025

To address the challenge of structural modeling and key information identification in zero-shot extractive summarization of long documents using large language models (LLMs), this paper proposes a training-free, graph-structured, three-stage prompting framework: neighborhood-aware prompting, centrality-aware prompting, and centrality-guided dynamic masking. It is the first work to integrate sentence-level graph construction and PageRank-based centrality analysis into zero-shot LLM inference, jointly enabling structural awareness and input compression optimization. Evaluated on benchmarks including ArXiv, the method achieves absolute improvements of 19.2 and 9.7 points in FactCC and SummaC-ZS scores, respectively—outperforming all unsupervised baselines significantly. Crucially, it preserves the zero-shot setting and requires no fine-tuning, maintaining conceptual and implementation simplicity while delivering state-of-the-art performance.

0 citationsRead paper

Recent publications

Latest Papers

Enhancing Distributed Authorization With Lagrange Interpolation And Attribute-Based Encryption

Dec 25, 2025

To address the high authorization overhead and key distribution bottlenecks in ciphertext data access within distributed environments, this paper proposes a lightweight access control scheme integrating attribute-based encryption (ABE) with an optimized key sharing mechanism. Methodologically, it innovatively incorporates second-order Lagrange interpolation into Shamir’s secret sharing to enable low-overhead key reconstruction, and synergistically combines involution-based stream ciphers with ABE to achieve both efficient decryption and fine-grained policy-based authorization. The scheme is formally proven secure against collusion attacks and provides forward secrecy. Experimental evaluation demonstrates a 37% reduction in end-to-end encryption/decryption latency, a 42% decrease in server-side computational overhead, and a 2.1× improvement in system throughput compared to state-of-the-art approaches.

0 citationsRead paper

StrucSum: Graph-Structured Reasoning for Long Document Extractive Summarization with LLMs

May 29, 2025

To address the challenge of structural modeling and key information identification in zero-shot extractive summarization of long documents using large language models (LLMs), this paper proposes a training-free, graph-structured, three-stage prompting framework: neighborhood-aware prompting, centrality-aware prompting, and centrality-guided dynamic masking. It is the first work to integrate sentence-level graph construction and PageRank-based centrality analysis into zero-shot LLM inference, jointly enabling structural awareness and input compression optimization. Evaluated on benchmarks including ArXiv, the method achieves absolute improvements of 19.2 and 9.7 points in FactCC and SummaC-ZS scores, respectively—outperforming all unsupervised baselines significantly. Crucially, it preserves the zero-shot setting and requires no fine-tuning, maintaining conceptual and implementation simplicity while delivering state-of-the-art performance.

0 citationsRead paper