Shengyi Jiang
Current Organizations / Institutes
The organizations/institutes from the scholar's latest article
Shengyi Jiang has a total of 77 co-authors. Here are the top co-authors by number of shared publications:
- Nankai Lin
School of Information Science and Technology, Guangdong University of Foreign Studies, Guangzhou, China
14 shared publications Nankai Lin
6 shared publicationsXiaotian Lin
School of Information Science and Technology Guangdong University of Foreign Studies Guangzhou Guangdong China
5 shared publications- Lihua Wang
Faculty of Data Science City University of Macau Macau SAR China
4 shared publications Sihui Fu
School of Information Science and Technology Guangdong University of Foreign Studies 510000 Guangzhou China
4 shared publicationsYingwen Fu
School of Information Science and Technology Guangdong University of Foreign Studies Guangzhou Guangdong China
4 shared publications
Latest Publications
- 11 May 2026
- Published by SAGE Publications in Intelligent Data Analysis: An International Journal
IN-LAND: Improving large language models’ performance on abbreviation disambiguation
Abbreviation disambiguation (AD) aims to select the most appropriate definition from a set of candidates. Large language models (LLMs) can perform various tasks without specific fine-tuning by using demonstration examples of in-context learning (ICL). However, significant challenges lie in harnessing the capabilities of LLMs for the AD task by designing effective ICL techniques. In this paper, we propose an IN -context- L earning-based A bbreviatio N D isambiguation framework ( IN-LAND ). IN-LAND introduces a novel mechanism for selecting demonstration examples in AD tasks, evaluating examples based on three criteria: strict abbreviation similarity, loose abbreviation similarity, and sentence semantic similarity. By selecting optimal training samples, this framework aims to improve the inference capabilities of LLMs. Our experimental results demonstrate the efficacy of IN-LAND in French, Legal English, and Malay. Compared to the Random K-shot ICL baseline, our method achieves consistent improvements across the average performance of four different LLMs. The impact is particularly notable in less commonly spoken languages. For instance, in Malay, our method improves the average macro F1 score and accuracy by 7.11% and 9.19%, respectively. In French, the increases are 2.69% for the macro F1 score and 3.17% for accuracy. Similarly, for Legal English, our approach enhances the average accuracy rate by 1.04% while maintaining competitive performance in macro F1 score. These results underscore the robust capabilities of LLMs and their ability to generalize effectively across various languages.
- 11 May 2026
- Published by SAGE Publications in Intelligent Data Analysis: An International Journal
IN-LAND: Improving large language models’ performance on abbreviation disambiguation
Abbreviation disambiguation (AD) aims to select the most appropriate definition from a set of candidates. Large language models (LLMs) can perform various tasks without specific fine-tuning by using demonstration examples of in-context learning (ICL). However, significant challenges lie in harnessing the capabilities of LLMs for the AD task by designing effective ICL techniques. In this paper, we propose an IN -context- L earning-based A bbreviatio N D isambiguation framework ( IN-LAND ). IN-LAND introduces a novel mechanism for selecting demonstration examples in AD tasks, evaluating examples based on three criteria: strict abbreviation similarity, loose abbreviation similarity, and sentence semantic similarity. By selecting optimal training samples, this framework aims to improve the inference capabilities of LLMs. Our experimental results demonstrate the efficacy of IN-LAND in French, Legal English, and Malay. Compared to the Random K-shot ICL baseline, our method achieves consistent improvements across the average performance of four different LLMs. The impact is particularly notable in less commonly spoken languages. For instance, in Malay, our method improves the average macro F1 score and accuracy by 7.11% and 9.19%, respectively. In French, the increases are 2.69% for the macro F1 score and 3.17% for accuracy. Similarly, for Legal English, our approach enhances the average accuracy rate by 1.04% while maintaining competitive performance in macro F1 score. These results underscore the robust capabilities of LLMs and their ability to generalize effectively across various languages.
- 8 March 2026
- Published by Elsevier in Expert Systems with Applications
- Vol. 319
Advancing LLMs for Chinese semantic error correction: Example selection and re-scoring
- 8 March 2026
- Published by Elsevier in Expert Systems with Applications
- Vol. 319
Advancing LLMs for Chinese semantic error correction: Example selection and re-scoring
- 28 February 2026
- Published by Elsevier in Information Fusion
- Vol. 133
Fusing representation spaces: A projected-fusion approach to contrastive learning
- 28 February 2026
- Published by Elsevier in Information Fusion
- Vol. 133
Fusing representation spaces: A projected-fusion approach to contrastive learning
- Open Access
- 4 February 2026
- Published by MDPI AG in Preprints.org
A Multi-Source Domain Transfer Fault Diagnosis Method Based on Graph Embedding and Adaptive Classifier
To address the prevalent domain inadaptation issue in multi-source domain transfer fault diagnosis, this study proposes a novel model integrating graph embedding technology and adaptive classifiers. The framework combines multi-statistic multi-scale permutation entropy methods to simultaneously optimize projection and classifier performance, thereby enhancing generalization capabilities and diagnostic accuracy. During model construction, a weighted non-parametric maximum mean difference approach is employed to quantify and minimize differences in edge distributions and conditional distributions between source and target domains. Through similarity weight allocation, norm adjustment, and row sparsity learning strategies, the model effectively reduces interference from irrelevant samples. Additionally, by embedding category information and domain manifold structures, the model ensures classification validity of mapped features. Using labeled source domain data and unlabeled target domain data, the study develops a structure-aware adaptive classifier that narrows distribution differences while preserving manifold consistency, significantly improving diagnostic performance under variable operating conditions. Experimental results demonstrate that this multi-source domain transfer model achieves higher accuracy than single-domain approaches. Notably, in rotating machinery variable operating scenarios, the proposed method outperforms existing techniques in both precision and generalization capabilities.
- 4 February 2026
- Published by MDPI AG in Preprints.org
Open AccessA Multi-Source Domain Transfer Fault Diagnosis Method Based on Graph Embedding and Adaptive Classifier
To address the prevalent domain inadaptation issue in multi-source domain transfer fault diagnosis, this study proposes a novel model integrating graph embedding technology and adaptive classifiers. The framework combines multi-statistic multi-scale permutation entropy methods to simultaneously optimize projection and classifier performance, thereby enhancing generalization capabilities and diagnostic accuracy. During model construction, a weighted non-parametric maximum mean difference approach is employed to quantify and minimize differences in edge distributions and conditional distributions between source and target domains. Through similarity weight allocation, norm adjustment, and row sparsity learning strategies, the model effectively reduces interference from irrelevant samples. Additionally, by embedding category information and domain manifold structures, the model ensures classification validity of mapped features. Using labeled source domain data and unlabeled target domain data, the study develops a structure-aware adaptive classifier that narrows distribution differences while preserving manifold consistency, significantly improving diagnostic performance under variable operating conditions. Experimental results demonstrate that this multi-source domain transfer model achieves higher accuracy than single-domain approaches. Notably, in rotating machinery variable operating scenarios, the proposed method outperforms existing techniques in both precision and generalization capabilities.
- 1 February 2026
- Published by Wiley in Transactions in GIS
- Vol. 30 (1)
Mining the Sense of Place in Chinese Cities Through Pop Music and LLMs
Sense of place is a significant topic in human geography, playing a critical role in cultural identity, urban planning, environmental protection, and sustainable tourism development. However, current research often relies on questionnaires and interviews that focus on specific communities or cities. This approach has limitations in covering a broader range of cities and does not support automated information extraction. To address these gaps, we propose an analytical framework that uses large language models (LLMs). First, we collect theme songs (pop music) and their associated comment data for 301 cities via NetEase Cloud Music, covering most prefecture‐level cities in China. Next, we develop LLM‐driven models, including a lyric similarity analysis model, a sentiment classification model, and a City DNA Extractor (focusing on landmarks, food, and cultural activities). In addition, we construct a city network to explore the relationships between cities. Finally, using these models and the constructed network, we analyze the sense of place both at a national scale and at the city level. Our study demonstrates the potential of music data and LLMs in sense‐of‐place research, revealing strong cultural identities and a deep sense of nostalgia among young people toward their hometowns, while offering a novel methodology and perspective for exploring this concept.
- 1 February 2026
- Published by Wiley in Transactions in GIS
- Vol. 30 (1)
Mining the Sense of Place in Chinese Cities Through Pop Music and LLMs
Sense of place is a significant topic in human geography, playing a critical role in cultural identity, urban planning, environmental protection, and sustainable tourism development. However, current research often relies on questionnaires and interviews that focus on specific communities or cities. This approach has limitations in covering a broader range of cities and does not support automated information extraction. To address these gaps, we propose an analytical framework that uses large language models (LLMs). First, we collect theme songs (pop music) and their associated comment data for 301 cities via NetEase Cloud Music, covering most prefecture‐level cities in China. Next, we develop LLM‐driven models, including a lyric similarity analysis model, a sentiment classification model, and a City DNA Extractor (focusing on landmarks, food, and cultural activities). In addition, we construct a city network to explore the relationships between cities. Finally, using these models and the constructed network, we analyze the sense of place both at a national scale and at the city level. Our study demonstrates the potential of music data and LLMs in sense‐of‐place research, revealing strong cultural identities and a deep sense of nostalgia among young people toward their hometowns, while offering a novel methodology and perspective for exploring this concept.
- 5 June 2025
- Published by Elsevier in International Journal of Disaster Risk Reduction
- Vol. 126
Government risk communication and response networks: Insights from typhoon news in China
- 5 June 2025
- Published by Elsevier in International Journal of Disaster Risk Reduction
- Vol. 126
Government risk communication and response networks: Insights from typhoon news in China
- 22 May 2025
- Published by Springer Nature
- p. 125-139
Character-Level Generative Network for Vietnamese Spelling Error Correction
- 22 May 2025
- Published by Springer Nature
- p. 125-139
Character-Level Generative Network for Vietnamese Spelling Error Correction
- Open Access
- 20 January 2025
- Published by Association for Computing Machinery (ACM) in ACM Transactions on Asian and Low-Resource Language Information Processing
- Vol. 24 (1) , 1-15
A Simple Yet Effective Corpus Construction Framework for Indonesian Grammatical Error Correction
Currently, the majority of research in grammatical error correction (GEC) is concentrated on universal languages, such as English and Chinese. Many low-resource languages lack accessible evaluation corpora. How to efficiently construct high-quality evaluation corpora for GEC in low-resource languages has become a significant challenge. To fill these gaps, in this article, we present a framework for constructing GEC corpora. Specifically, we focus on Indonesian as our research language and construct an evaluation corpus for Indonesian GEC using the proposed framework, addressing the limitations of existing evaluation corpora in Indonesian. Furthermore, we investigate the feasibility of utilizing existing large language models (LLMs), such as GPT-3.5-Turbo and GPT-4, to streamline corpus annotation efforts in GEC tasks. The results demonstrate significant potential for enhancing the performance of LLMs in low-resource language settings. Our code and corpus can be obtained from https://github.com/GKLMIP/GEC-Construction-Framework .
- 20 January 2025
- Published by Association for Computing Machinery (ACM) in ACM Transactions on Asian and Low-Resource Language Information Processing
- Vol. 24 (1) , 1-15
Open AccessA Simple Yet Effective Corpus Construction Framework for Indonesian Grammatical Error Correction
Currently, the majority of research in grammatical error correction (GEC) is concentrated on universal languages, such as English and Chinese. Many low-resource languages lack accessible evaluation corpora. How to efficiently construct high-quality evaluation corpora for GEC in low-resource languages has become a significant challenge. To fill these gaps, in this article, we present a framework for constructing GEC corpora. Specifically, we focus on Indonesian as our research language and construct an evaluation corpus for Indonesian GEC using the proposed framework, addressing the limitations of existing evaluation corpora in Indonesian. Furthermore, we investigate the feasibility of utilizing existing large language models (LLMs), such as GPT-3.5-Turbo and GPT-4, to streamline corpus annotation efforts in GEC tasks. The results demonstrate significant potential for enhancing the performance of LLMs in low-resource language settings. Our code and corpus can be obtained from https://github.com/GKLMIP/GEC-Construction-Framework .
- 14 November 2024
- Published by Elsevier in Computer Speech & Language
- Vol. 91
Corpus and unsupervised benchmark: Towards Tagalog grammatical error correction
- 14 November 2024
- Published by Elsevier in Computer Speech & Language
- Vol. 91
Corpus and unsupervised benchmark: Towards Tagalog grammatical error correction