Shuqin Li
Current Organizations / Institutes
The organizations/institutes from the scholar's latest article
Shuqin Li has a total of 20 co-authors. Here are the top co-authors by number of shared publications:
- Yuntao Shi
School of electrical and control engineering, North China University of Technology, Beijing 100144, China
7 shared publications - Jie Li
School of Artificial Intelligence and Computer, North China University of Technology, Beijing 100144, China
6 shared publications - Meng Zhou
School of electrical and control engineering, North China University of Technology, Beijing 100144, China
4 shared publications - Yu Ding
School of electrical and control engineering, North China University of Technology, Beijing 100144, China
3 shared publications - Jie Li
North China University of Technology, 5 Jinyuanzhuang Road, Beijing, 100144, PR China
2 shared publications - Wei Guo
School of electrical and control engineering, North China University of Technology, Beijing 100144, China
2 shared publications
Latest Publications
- Open Access
- 25 April 2026
- Published by MDPI AG in Electronics
- Vol. 15 (9) , 1826
Construction Method of Campus Safety Knowledge Graph Based on Retrieval-Augmented
LLMs suffer from the hallucination problem in knowledge extraction within the campus safety domain, which impairs the accuracy of extracting campus safety entities and relationships. To address this issue, this paper proposes a retrieval-augmented method for constructing a campus safety knowledge graph. The proposed methodology incorporates rule-based text segmentation of campus safety texts, pre-entity extraction from campus safety data, and similarity checks on each pre-entity to achieve entity disambiguation. Next, it utilizes a general LLM for inter-entity relationship extraction to build an initial campus safety knowledge graph. Finally, the method performs vector similarity retrieval on the campus safety data and the initial knowledge graph to obtain relevant text chunks and subgraphs, which are then input into the LLM via prompts to construct a more comprehensive campus safety knowledge graph. On the MINE dataset, the proposed method achieved 71.55% entity faithfulness; on the campus safety dataset, it achieved 79.85% entity faithfulness. These results are 8.62% and 8.30% higher than those of KGGen and GraphRAG, respectively, demonstrating that the proposed method enhances the accuracy of extracting campus safety entities and relationships from LLMs.
- 25 April 2026
- Published by MDPI AG in Electronics
- Vol. 15 (9) , 1826
Open AccessConstruction Method of Campus Safety Knowledge Graph Based on Retrieval-Augmented
LLMs suffer from the hallucination problem in knowledge extraction within the campus safety domain, which impairs the accuracy of extracting campus safety entities and relationships. To address this issue, this paper proposes a retrieval-augmented method for constructing a campus safety knowledge graph. The proposed methodology incorporates rule-based text segmentation of campus safety texts, pre-entity extraction from campus safety data, and similarity checks on each pre-entity to achieve entity disambiguation. Next, it utilizes a general LLM for inter-entity relationship extraction to build an initial campus safety knowledge graph. Finally, the method performs vector similarity retrieval on the campus safety data and the initial knowledge graph to obtain relevant text chunks and subgraphs, which are then input into the LLM via prompts to construct a more comprehensive campus safety knowledge graph. On the MINE dataset, the proposed method achieved 71.55% entity faithfulness; on the campus safety dataset, it achieved 79.85% entity faithfulness. These results are 8.62% and 8.30% higher than those of KGGen and GraphRAG, respectively, demonstrating that the proposed method enhances the accuracy of extracting campus safety entities and relationships from LLMs.
- Open Access
- 27 March 2026
- Published by Association for Computing Machinery (ACM)
- p. 45-49
Research on Multi-Source Data Preprocessing Methods for Depression Recognition
To address the limitations of existing depression datasets, including large inter-class variability and the lack of unified standardization that hinder recognition performance, this study investigates multimodal data fusion for depression detection. Based on the DAIC-WOZ dataset and guided by PHQ-8 scores together with DSM-5 diagnostic criteria, all samples are categorized into three classes: Non-depressed (122 cases), Situational depression (53 cases), and Major depressive disorder (14 cases). Unified preprocessing procedures are applied across modalities, including audio processing (16 kHz resampling, denoising, STFT and Mel-spectrogram conversion), facial data processing (MTCNN-based face detection and Dlib keypoint extraction mapped to Gaussian heatmaps), and text processing (cleaning and normalization). This standardized preprocessing pipeline establishes a reliable multimodal data foundation, reduces cross-modal heterogeneity, and facilitates subsequent depression recognition and performance improvement.
- 27 March 2026
- Published by Association for Computing Machinery (ACM)
- p. 45-49
Open AccessResearch on Multi-Source Data Preprocessing Methods for Depression Recognition
To address the limitations of existing depression datasets, including large inter-class variability and the lack of unified standardization that hinder recognition performance, this study investigates multimodal data fusion for depression detection. Based on the DAIC-WOZ dataset and guided by PHQ-8 scores together with DSM-5 diagnostic criteria, all samples are categorized into three classes: Non-depressed (122 cases), Situational depression (53 cases), and Major depressive disorder (14 cases). Unified preprocessing procedures are applied across modalities, including audio processing (16 kHz resampling, denoising, STFT and Mel-spectrogram conversion), facial data processing (MTCNN-based face detection and Dlib keypoint extraction mapped to Gaussian heatmaps), and text processing (cleaning and normalization). This standardized preprocessing pipeline establishes a reliable multimodal data foundation, reduces cross-modal heterogeneity, and facilitates subsequent depression recognition and performance improvement.
- Open Access
- 1 March 2026
- Published by MDPI AG in Algorithms
- Vol. 19 (3) , 182
MACFormer: Multi-Dimensional Attention and Composite Loss Former for Enhancing Few-Shot Image Classification
Addressing challenges in few-shot image classification, this study introduces the Multi-Dimensional Attention and Composite Loss Former, a meta-learning model built on a Residual Network-12 backbone. The model incorporates multi-dimensional attention mechanisms and is trained with a composite loss function applied across the entire architecture. It enhances feature extraction by dynamically focusing on critical local and global information, while the composite loss optimizes classification accuracy, emphasizes hard samples, suppresses overfitting, and promotes intra-class feature compactness. Comprehensive experiments conducted on the miniImageNet and tieredImageNet datasets demonstrate that the proposed model achieves superior performance in both meta-training and meta-testing stages compared to existing benchmarks, effectively validating its robustness and generalization capabilities in few-shot learning tasks.
- 1 March 2026
- Published by MDPI AG in Algorithms
- Vol. 19 (3) , 182
Open AccessMACFormer: Multi-Dimensional Attention and Composite Loss Former for Enhancing Few-Shot Image Classification
Addressing challenges in few-shot image classification, this study introduces the Multi-Dimensional Attention and Composite Loss Former, a meta-learning model built on a Residual Network-12 backbone. The model incorporates multi-dimensional attention mechanisms and is trained with a composite loss function applied across the entire architecture. It enhances feature extraction by dynamically focusing on critical local and global information, while the composite loss optimizes classification accuracy, emphasizes hard samples, suppresses overfitting, and promotes intra-class feature compactness. Comprehensive experiments conducted on the miniImageNet and tieredImageNet datasets demonstrate that the proposed model achieves superior performance in both meta-training and meta-testing stages compared to existing benchmarks, effectively validating its robustness and generalization capabilities in few-shot learning tasks.
- Open Access
- 14 November 2025
- Published by Association for Computing Machinery (ACM)
- p. 256-260
A Depression Detection Method Based on Emotion-Topic Guided In-Context Learning
- 14 November 2025
- Published by Association for Computing Machinery (ACM)
- p. 256-260
Open AccessA Depression Detection Method Based on Emotion-Topic Guided In-Context Learning
- Open Access
Integrating Pose Features and Cross-Relationship Learning for Human–Object Interaction Detection
Background: The main challenge in human–object interaction detection (HOI) is how to accurately reason about ambiguous, complex, and difficult to recognize interactions. The model structure of the existing methods is relatively single, and the image input may be occluded and cannot be accurately recognized. Methods: In this paper, we design a Pose-Aware Interaction Network (PAIN) based on transformer architecture and human posture to address these issues through two innovations: A new feature fusion method is proposed, which fuses human pose features and image features early before the encoder to improve the feature expression ability, and the individual motion-related features are additionally strengthened by adding to the human branch; the Cross-Attention Relationship fusion Module (CARM) better fuses the three-branch output and captures the detailed relationship information of HOI. Results: The proposed method achieves 64.51%, 66.42% on the public dataset V-COCO and 30.83% AP on HICO-DET, which can recognize HOI instances more accurately.Open AccessIntegrating Pose Features and Cross-Relationship Learning for Human–Object Interaction Detection
Background: The main challenge in human–object interaction detection (HOI) is how to accurately reason about ambiguous, complex, and difficult to recognize interactions. The model structure of the existing methods is relatively single, and the image input may be occluded and cannot be accurately recognized. Methods: In this paper, we design a Pose-Aware Interaction Network (PAIN) based on transformer architecture and human posture to address these issues through two innovations: A new feature fusion method is proposed, which fuses human pose features and image features early before the encoder to improve the feature expression ability, and the individual motion-related features are additionally strengthened by adding to the human branch; the Cross-Attention Relationship fusion Module (CARM) better fuses the three-branch output and captures the detailed relationship information of HOI. Results: The proposed method achieves 64.51%, 66.42% on the public dataset V-COCO and 30.83% AP on HICO-DET, which can recognize HOI instances more accurately. - Open Access
- 30 April 2024
- Published by Faculty of Transport and Traffic Sciences in Promet - Traffic&Transportation
- Vol. 36 (2) , 279-293
Analysis of Beijing Traffic Violations Based on the BERT-CRF Model
Traffic violations are a major cause of traffic accidents, yet current research falls short in comprehensively analysing these violations and the named entity method fails to extract the name of traffic violation events from records, thereby lacking in providing guidance for managing urban traffic violations. By expanding the People’s Daily dataset from 71,456 words to 95,291 words, the BERT-CRF (Bidirectional Encoder Representations from Transformers-Conditional Random Field) model achieves an accuracy rate of 88.53%, a recall rate of 92.90% and an F1 score of 90.66%, successfully identifying event, time and location named entities within traffic violations. The data of traffic violations is then enhanced through forward geocoding and the Bayesian formula, and traffic violations are analysed from time, space, administrative region, gender and weather, to provide support for the dynamic allocation of law enforcement forces on traffic scenes and the precise management oftraffic violations.
- 30 April 2024
- Published by Faculty of Transport and Traffic Sciences in Promet - Traffic&Transportation
- Vol. 36 (2) , 279-293
Open AccessAnalysis of Beijing Traffic Violations Based on the BERT-CRF Model
Traffic violations are a major cause of traffic accidents, yet current research falls short in comprehensively analysing these violations and the named entity method fails to extract the name of traffic violation events from records, thereby lacking in providing guidance for managing urban traffic violations. By expanding the People’s Daily dataset from 71,456 words to 95,291 words, the BERT-CRF (Bidirectional Encoder Representations from Transformers-Conditional Random Field) model achieves an accuracy rate of 88.53%, a recall rate of 92.90% and an F1 score of 90.66%, successfully identifying event, time and location named entities within traffic violations. The data of traffic violations is then enhanced through forward geocoding and the Bayesian formula, and traffic violations are analysed from time, space, administrative region, gender and weather, to provide support for the dynamic allocation of law enforcement forces on traffic scenes and the precise management oftraffic violations.
- Open Access
- 11 April 2024
- Published by MDPI AG in Information
- Vol. 15 (4) , 215
LighterFace Model for Community Face Detection and Recognition
This research proposes a face detection algorithm named LighterFace, which is aimed at enhancing detection speed to meet the demands of real-time community applications. Two pre-trained convolutional neural networks are combined, namely Cross Stage Partial Network (CSPNet), and ShuffleNetv2. Connecting the optimized network with Global Attention Mechanism (GAMAttention) extends the model to compensate for the accuracy loss caused by optimizing the network structure. Additionally, the learning rate of the detection model is dynamically updated using the cosine annealing method, which enhances the convergence speed of the model during training. This paper analyzes the training of the LighterFace model on the WiderFace dataset and a custom community dataset, aiming to classify faces in real-life community settings. Compared to the mainstream YOLOv5 model, LighterFace demonstrates a significant reduction in computational demands by 85.4% while achieving a 66.3% increase in detection speed and attaining a 90.6% accuracy in face detection. It is worth noting that LighterFace generates high-quality cropped face images, providing valuable inputs for subsequent face recognition models such as DeepID. Additionally, the LighterFace model is specifically designed to run on edge devices with lower computational capabilities. Its real-time performance on a Raspberry Pi 3B+ validates the results.
- 11 April 2024
- Published by MDPI AG in Information
- Vol. 15 (4) , 215
Open AccessLighterFace Model for Community Face Detection and Recognition
This research proposes a face detection algorithm named LighterFace, which is aimed at enhancing detection speed to meet the demands of real-time community applications. Two pre-trained convolutional neural networks are combined, namely Cross Stage Partial Network (CSPNet), and ShuffleNetv2. Connecting the optimized network with Global Attention Mechanism (GAMAttention) extends the model to compensate for the accuracy loss caused by optimizing the network structure. Additionally, the learning rate of the detection model is dynamically updated using the cosine annealing method, which enhances the convergence speed of the model during training. This paper analyzes the training of the LighterFace model on the WiderFace dataset and a custom community dataset, aiming to classify faces in real-life community settings. Compared to the mainstream YOLOv5 model, LighterFace demonstrates a significant reduction in computational demands by 85.4% while achieving a 66.3% increase in detection speed and attaining a 90.6% accuracy in face detection. It is worth noting that LighterFace generates high-quality cropped face images, providing valuable inputs for subsequent face recognition models such as DeepID. Additionally, the LighterFace model is specifically designed to run on edge devices with lower computational capabilities. Its real-time performance on a Raspberry Pi 3B+ validates the results.
- Open Access
- 29 March 2024
- Published by MDPI AG in Information
- Vol. 15 (4) , 188
S-YOLOv5: A Lightweight Model for Detecting Objects Thrown from Tall Buildings in Communities
Objects thrown from tall buildings in communities are characterized by their small size, inconspicuous features, and high speed. Existing algorithms for detecting such objects face challenges, including excessive parameters, overly complex models that are difficult to implement, and insufficient detection accuracy. This study proposes a lightweight detection model for objects thrown from tall buildings in communities, named S-YOLOv5, to address these issues. The model is based on the YOLOv5 algorithm, and a lightweight convolutional neural network, Enhanced ShuffleNet (ESNet), is chosen as the backbone network to extract image features. On this basis, the initial stage of the backbone network is enhanced and the simplified attention module (SimAM) attention mechanism is added to utilize the rich position information and contour information in the shallow feature map to improve the detection of small targets. For feature fusion, the sparsely connected Path Aggregation Network (SCPANet) module is designed to use sparsely connected convolution (SCConv) instead of the regular convolution of the Path Aggregation Network (PANet) to fuse features efficiently. In addition, the model uses the normalized Wasserstein distance (NWD) loss function to reduce the sensitivity of positional bias. The accuracy of the model is further improved. Test results from the self-built objects thrown from tall buildings dataset show that S-YOLOv5 can detect objects thrown from tall buildings quickly and accurately, with an accuracy of 90.2% and a detection rate of 34.1 Fps/s. Compared with the original YOLOv5 model, the parameters are reduced by 87.3%, and the accuracy and rate are improved by 0.8% and 63%, respectively.
- 29 March 2024
- Published by MDPI AG in Information
- Vol. 15 (4) , 188
Open AccessS-YOLOv5: A Lightweight Model for Detecting Objects Thrown from Tall Buildings in Communities
Objects thrown from tall buildings in communities are characterized by their small size, inconspicuous features, and high speed. Existing algorithms for detecting such objects face challenges, including excessive parameters, overly complex models that are difficult to implement, and insufficient detection accuracy. This study proposes a lightweight detection model for objects thrown from tall buildings in communities, named S-YOLOv5, to address these issues. The model is based on the YOLOv5 algorithm, and a lightweight convolutional neural network, Enhanced ShuffleNet (ESNet), is chosen as the backbone network to extract image features. On this basis, the initial stage of the backbone network is enhanced and the simplified attention module (SimAM) attention mechanism is added to utilize the rich position information and contour information in the shallow feature map to improve the detection of small targets. For feature fusion, the sparsely connected Path Aggregation Network (SCPANet) module is designed to use sparsely connected convolution (SCConv) instead of the regular convolution of the Path Aggregation Network (PANet) to fuse features efficiently. In addition, the model uses the normalized Wasserstein distance (NWD) loss function to reduce the sensitivity of positional bias. The accuracy of the model is further improved. Test results from the self-built objects thrown from tall buildings dataset show that S-YOLOv5 can detect objects thrown from tall buildings quickly and accurately, with an accuracy of 90.2% and a detection rate of 34.1 Fps/s. Compared with the original YOLOv5 model, the parameters are reduced by 87.3%, and the accuracy and rate are improved by 0.8% and 63%, respectively.
- 2 October 2023
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Transactions on Artificial Intelligence
- Vol. 5 (5) , 2217-2226
Temporal Knowledge Graph for Food Risk Prediction
Accurate prediction of food-safety risk is an effective measure to improve food risk prevention. The complex and continuous characteristics of food safety influence various factors, thus, a food-safety risk prediction model based on a temporal knowledge graph is proposed in this paper. First, a food-safety risk dataset is constructed by collecting food supervision and management sampling data used in daily life from 2018 to 2021 from the State Administration for Market Regulation. The dataset contains five categories: fruits, vegetables, meat, aquatic products, and dairy products. Then, a novel food-safety temporal knowledge graph is designed based on the proposed index system because food-safety data have temporal characteristics. A temporal knowledge graph network is proposed to build a food safety risk prediction model, which comprises historical learning and generation methods. The proposed food-safety temporal knowledge graph can predict the food risk level and types of hazardous substances in a certain period. Finally, the comparative experiments on the food safety dataset constructed in this article showed that the proposed model achieved the accuracy of 86.15%, the Mean Reciprocal Rank (MRR) of 88.64%, and the recall of 85.13%. This demonstrated that the proposed food safety risk prediction method based on temporal knowledge graph networks has higher accuracy and stronger interpretability compared to some existing data prediction methods
- 2 October 2023
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Transactions on Artificial Intelligence
- Vol. 5 (5) , 2217-2226
Temporal Knowledge Graph for Food Risk Prediction
Accurate prediction of food-safety risk is an effective measure to improve food risk prevention. The complex and continuous characteristics of food safety influence various factors, thus, a food-safety risk prediction model based on a temporal knowledge graph is proposed in this paper. First, a food-safety risk dataset is constructed by collecting food supervision and management sampling data used in daily life from 2018 to 2021 from the State Administration for Market Regulation. The dataset contains five categories: fruits, vegetables, meat, aquatic products, and dairy products. Then, a novel food-safety temporal knowledge graph is designed based on the proposed index system because food-safety data have temporal characteristics. A temporal knowledge graph network is proposed to build a food safety risk prediction model, which comprises historical learning and generation methods. The proposed food-safety temporal knowledge graph can predict the food risk level and types of hazardous substances in a certain period. Finally, the comparative experiments on the food safety dataset constructed in this article showed that the proposed model achieved the accuracy of 86.15%, the Mean Reciprocal Rank (MRR) of 88.64%, and the recall of 85.13%. This demonstrated that the proposed food safety risk prediction method based on temporal knowledge graph networks has higher accuracy and stronger interpretability compared to some existing data prediction methods
- 17 February 2022
- Published by Elsevier in Journal of Food Engineering
- Vol. 323
Heterogeneous graph attention network for food safety risk prediction
- 17 February 2022
- Published by Elsevier in Journal of Food Engineering
- Vol. 323
Heterogeneous graph attention network for food safety risk prediction