From 454553e465421d770f6e18de55285294d3a7fedb Mon Sep 17 00:00:00 2001 From: caoyatao555 <201615625@stu.ncwu.edu.cn> Date: Wed, 14 Aug 2024 11:25:51 +0800 Subject: [PATCH] Update mlic24.bib The order of the author's surname and given name was incorrect, so it has been updated. --- mlic24.bib | 94 +++++++++++++++++++++++++++--------------------------- 1 file changed, 47 insertions(+), 47 deletions(-) diff --git a/mlic24.bib b/mlic24.bib index 84f3be7..1b232dc 100644 --- a/mlic24.bib +++ b/mlic24.bib @@ -17,49 +17,49 @@ @Proceedings{MLIC-2024 @InProceedings{mlic24-01, title = {Event Extraction in Complex Sentences Based on Dependency Parsing and Longformer}, - author = {Lin, Li and Ziyang, Chen and Shuxing, Liao and Yibin, Du and Hongxiao, Wu and Zhihao, Li}, + author = {Li, Lin and Chen, Ziyang and Liao, Shuxing and Du, Yibin and Wu, Hongxiao and Li, Zhihao}, pages = {1-7}, abstract = {Event extraction involves the identification and extraction of specific event-related information from a large corpus of textual data. In recent years, the introduction of pre-trained models has significantly enhanced the ability of these models to comprehend the semantics of sentences, leading to continuous advancements in event extraction methods. However, when it comes to long and complex sentences, these models have shown limited performance. This limitation can be attributed to the intricate structures of such sentences, which hinder the models' ability to grasp their semantic meaning. To tackle this challenge, we propose a novel model that combines dependency analysis tools with the Longformer pre-trained model. By effectively analyzing the structures of complex sentences, our model aims to enhance the semantic understanding of these sentences. Experimental results using the ACE2005 dataset demonstrate the improved performance of our model in event extraction for complex sentences.} } @InProceedings{mlic24-02, title = {Biomedical Relation Extraction Based on Deep Transfer Learning and Prompt Learning}, - author = {Rongrong, Ma and Huan, Liu}, + author = {Ma, Rongrong and Liu, Huan}, pages = {8-15}, abstract = {Biomedical Relationship Extraction (BioRE) is an important task for electronic health record mining and biomedical knowledge base construction. Due to the complexity of biomedical information in the literature, it is difficult to realize the construction of large-scale datasets. Although previous models showed good results in a fully supervised environment, they did not generalize well to the low resource situation common in this field. We propose a biomedical relationship extraction model based on deep transfer learning and prompt learning. We use deep neural network for transfer learning, using large-scale biomedical field data set (BioREx) training model. The trained model was applied to low-resource datasets in biomedical field, which can make the model to learn more rich knowledge, so as to alleviate the problem of insufficient training data. In addition, we used a prompt learning method by appending a prompt sentence describing the desired relationship label to the beginning of the input sentence, which can reduce the gap between the pretrained language model (PLM) and downstream tasks. Experiments show that using deep transfer learning and prompt learning can effectively improve the prediction results. Experiments on three BioRE benchmark datasets DrugProt, DDI and BC5CDR, F1 values are significantly improved compared to previous models.} } @InProceedings{mlic24-03, title = {SpcNet: Speaker Validation Model Based on Self-Calibrated Convolution}, - author = {Xia, Zhang and Guobo, Wu and Qian, Liu}, + author = {Zhang, Xia and Wu, Guobo and Liu, Qian}, pages = {16-24}, abstract = {The convolutional module-based speaker representation network has demonstrated outstanding performance in the speaker verification (SV) task and has now become one of the widely adopted network structures in this task field. There are some limitations to the convolution-based network structure, specifically with the fixed-size convolution kernel in the general convolution operation. This makes it difficult to capture long-range time-frequency and channel dependencies in speech features, limiting the network's ability to extract representations of the speaker. To overcome this issue, we have explored several alternative approaches. Firstly, we propose an enhanced selfcalibrating convolutional kernel that adaptively constructs long-range time-frequency and channel dependencies around each time-frequency position. This allows for the integration of richer information, significantly enhancing the network's capacity to learn representations. Secondly, we have made adjustments to the network structure to improve the extraction of speaker feature representations. We refer to this modified model as SpcNet. In this paper, our proposed SpcNet model has been experimented on the datasets VoxCeleb1 and VoxCeleb2. Comprehensive experiments show that the Equal Error Rate (EER) is significantly improved.} } @InProceedings{mlic24-04, title = {Predicting Student Performance with Graph Structure Density-Based Graph Neural Networks}, - author = {Xiaochen, Lai and Wentao, Hao and Zheng, Zhang and Xiaohan, Bai}, + author = {Lai, Xiaochen and Hao, Wentao and Zhang, Zheng and Bai, Xiaohan}, pages = {25-32}, abstract = {Educational data mining is a key area in data mining, focusing on predicting student performance. This research offers a new perspective for EDM, aiding educators in timely interventions based on performance predictions to reduce student failure and improve teaching quality. A novel approach: Graph Structure Density-based Sampling-Aggregation (GSDSA), is presented. GSDSA effectively reduces computational costs and improves training efficiency by limiting the number of neighbor nodes and using the Jaccard coefficient to measure node similarity. Moreover, it also considers relationships among students, employing the Pearson correlation coefficient to analyze none-graph-structured data, enhancing its processing capabilities. Experimental results demonstrate that GSDSA surpasses various graph neural networks and machine learning algorithms in classification accuracy.} } @InProceedings{mlic24-05, title = {Collaborative Learning with Curriculum Loss for Accurate and InterpretableWeakly Supervised Whole-Slide Image Classification Collaborative Learning with Curriculum Loss for Whole-Slide Image Classification}, - author = {Rui, Yang and Pei, Liu and Luping, Ji}, + author = {Yang, Rui and Liu, Pei and Ji, Luping}, pages = {33-39}, abstract = {Weakly-Supervised Learning (WSL) has been increasingly concerned in Whole-Slide Image (WSI) classification, meanwhile, an open question arises: could WSL-based models provide us with an accurate interpretation of their decisions? Although many research works have made exciting progress via building an Auxiliary Instance Branch (AIB) on a bag-level network, there are still two typical problems to be confronted with in training WSL-based AIB: i) an overwhelming influence of negative instances and ii) the inconsistent learning between bag-level network and AIB. To address them, this paper proposes collaborative learning with curriculum loss. This scheme, on one hand, provides a curriculum loss for optimizing AIB, to alleviate the first problem. Considering the knowledge reliability in WSL, this loss generalizes an original quality focal loss to WSL scenarios by curriculum instances. On the other hand, to overcome the second problem, this scheme trains a bag-level network under the supervision of AIB by a reversed curriculum loss, making both learn collaboratively. Comparative experiments prove that our scheme could often surpass existing ones in both accuracy and interpretability. Moreover, it is found that the knowledge reliability-inspired curriculum instance is a critical factor in bringing comprehensive improvements.} } @InProceedings{mlic24-06, title = {A Word Sense Disambiguation Method Based on Multiple Sense Graph}, - author = {Wenjun, Liu and Hailan, Wang and Xiping, Wang and Mengshu, Hou}, + author = {Liu, Wenjun and Wang, Hailan and Wang, Xiping and Hou, Mengshu}, pages = {40-47}, abstract = {Word Sense Disambiguation is a process to determine the best meaning of an ambiguous word according to its contextual semantic information. Many methods ofWord Sense Disambiguation cannot deal with polysemous words well because they only consider the meaning of the adjacent words before and after ambiguous words, and cannot consider the meaning of all words in the sentence globally. In order to solve the above problems, this paper proposes a word sense disambiguation method based on Multiple Sense Graph. This method applies the BERT model to generate word sense vectors, and globally considers the feature relationship between the ambiguous word and all words in the context. In addition, this method applies the PageRank algorithm to score the importance of each sense vector of the word, and the scoring results are sorted to obtain the best sense of the ambiguous word. The experimental results indicate that the proposed BERT-PageRank method improves the evaluation index compared with the other two semantic disambiguation methods. In summary, the proposed method improves the accuracy of word sense disambiguation to obtain the best word sense.} } @InProceedings{mlic24-07, title = {Thesis Reviewer Recommendation Based on Multi-Graph Neural Networks}, - author = {Tao, Wang and Peng, Fu and Jing, Zhang and Kaixuan, Chen}, + author = {Wang, Tao and Fu, Peng and Zhang, Jing and Chen, Kaixuan}, pages = {48-56}, abstract = {Thesis review is a crucial step in ensuring the quality of academic theses, and accurately recom-mending reviewers for theses is currently a problem that needs to be addressed. When reviewer information is incomplete, it is difficult to achieve good recommendation results. This paper pro-poses a Multi-Graph Neural Network algorithm for review reviewer recommendation in thesis re-view. Based on the keyword-reviewer bipartite graph, a graph neural network model is constructed. By utilizing graph neural networks, high-order relationships between reviews and keywords can be explored, enabling the discovery of reviews' implicit research interests and expanding their re-search interests to some extent. Additionally, incorporating keyword-keyword interaction graphs and review-review interaction graphs allows for information exchange operations in the two graphs separately, enhancing the representation of keywords and reviews. We conducted experiments on real thesis reviews and compared the proposed algorithm with other recommendation algorithms. The results show that the proposed algorithm achieves favorable results across various evaluation metrics, demonstrating the effectiveness of the algorithm presented in this paper. } @@ -67,7 +67,7 @@ @InProceedings{mlic24-07 @InProceedings{mlic24-08, title = {Multiscale Fusion and Boundary Refinement UNet Model for Lung Parenchyma Segmentation}, - author = {Hehe, Huang and Senyu, Wei and Chen, Zhao and Yu, Shi and Yuetao, Zhao and Yuan, Li and Qiuming, Zang and Jianhua, Xu}, + author = {Huang, Hehe and Wei, Senyu and Zhao, Chen and Shi, Yu and Zhao, Yuetao and Li, Yuan and Zang, Qiuming and Xu, Jianhua}, pages = {57-66}, abstract = {Methods based on automatic computer-aided systems for lung cancer diagnosis have gained popu-larity in recent years. Lung parenchyma segmentation technology plays an important role in these methods. To reduce the gradient loss, improve the feature utilization, and alleviate the difficulty of fully mining the contextual information in lung parenchyma segmentation, this study proposes a network model for lung parenchyma segmentation based on a Multiscale fusion and Boundary Refinement UNet (MBR–UNet) model. First, along the encoding path, the features are efficiently extracted by a Residual–Residual block module. Next, along the decoding path, a multiscale at-tention–spatial pyramid pooling module fully integrates the feature maps of different layers and sums the outputs of each layer of the decoding path for boundary refinement. Finally, the training model is optimized through a hybrid loss function. The proposed model is experimentally evaluated on the lung segmentation dataset of the Kaggle competition. The accuracy, Dice similarity coefficient, intersection ratio, and Hausdorff distance of the network segmentation are improved by 98.79\%, 97.35\%, 96.34\%, and 12.82 mm, respectively, from those of other segmentation methods. According to these results, the method can more precisely segment pulmonary parenchyma than the existing methods. } @@ -75,7 +75,7 @@ @InProceedings{mlic24-08 @InProceedings{mlic24-09, title = {Research on Features Extraction and Classification for Images based on Transformer Learning}, - author = {Chao, Wang}, + author = {Wang, Chao}, pages = {67-75}, abstract = {Image processing and analysis have become an essential method in many areas including medical impact, facial recognition, and social media analysis. With the rapid development of big data and artificial intelligence technology, especially the emergence of Transformer learning models, new methods have been brought to image feature extraction and classification. However, the existing transformer model limits the ability to handle variable-length sequences and understand complex sequence relationships. In this work, we propose a novel transformer-based framework that combines a self-attention mechanism and a multi-head attention technique to efficiently extract features from complex image data. In addition, we introduce an improved classifier that enables efficient image classification using extracted features. Our method takes into account not only the local features of the image but also the global relationships between different regions to achieve a more accurate representation of the features. We simulate our model with existing convolutional neural networks and other traditional machine learning methods in the public datasets including CIFAR-10 and MNIST. From our experimental results, we can observe that our transformer-learning-based framework shows significant performance improvement in image feature extraction and classifica-tion tasks. } @@ -83,7 +83,7 @@ @InProceedings{mlic24-09 @InProceedings{mlic24-10, title = {A Trust Evaluation Method with Collaborative Active Detection for D2D-enabled Networks}, - author = {Gang, Yang and Pan, Liu and Yan, Liu}, + author = {Yang, Gang and Liu, Pan and Liu, Yan}, pages = {76-83}, abstract = {Device-to-Device (D2D) communication emerges as a pivotal technology for enhancing the capacity and coverage of networks. Nonetheless, these networks are vulnerable to threats posed by malicious devices (MDs) that have the potential to tamper with transmitted content, thereby com-promising the network's reliability. The trust model is one of the effective methods to solve such internal attacks. Previous research mainly evaluated device trust based on social similarity and passive trust models. These methods make it difficult to obtain accurate evaluation results. Even though there are some active trust models, they have limitations such as high cost and limited scope of application. To solve the above problems, we propose a low-cost collaborative active detection trust evaluation method. In this method, the device first generates some smaller detection packets, verification codes and sends them directly to the trusted alliance party. Then, during each trust eval-uation process, the device determines the trust of nodes on the multi-hop path by actively sending these detection packets to the trusted alliance party. Experimental results show that, compared with existing strategies, our proposed strategy can achieve higher trust evaluation accuracy with lower energy consumption. } @@ -91,7 +91,7 @@ @InProceedings{mlic24-10 @InProceedings{mlic24-11, title = {A Truth Discovery Method with Trust-Aware Self-Supervised Model for Visual Crowdsensing}, - author = {Yan, Liu and Hang, Zhang and Pan, Liu and Gang, Yang}, + author = {Liu, Yan and Zhang, Hang and Liu, Pan and Yang, Gang}, pages = {84-93}, abstract = {Crowdsensing relies on truth discovery methods to obtain reliable data. However, existing truth dis-covery methods for Crowd Sensing systems face challenges in effectively evaluating the quality of uploaded image data by workers and defending against attacks by malicious workers. To overcome these issues, we introduce a truth discovery method with the trust-aware self-supervised discrim-inative (SSD) model for visual crowdsensing, namely TASSD. In the TASSD, we incorporate the worker's trustworthiness in the reweighting mechanism of the SSD model, thereby improving system robustness and reliability. Then, we designed a trust update method to accurately obtain the worker's trustworthiness. Experimental results demonstrate the superiority of our proposed TASSD over traditional anomaly detection methods, particularly in scenarios with high ratios of abnormal data. TASSD effectively addresses disguised malicious worker attacks that achieve high credibility. } @@ -99,7 +99,7 @@ @InProceedings{mlic24-11 @InProceedings{mlic24-12, title = {Construction Method for Knowledge Graph of Driving Behavior under Adverse Weather}, - author = {Yi, Tian and Zhangcai, Yin and Yiran, Chen}, + author = {Tian, Yi and Yin, Zhangcai and Chen, Yiran}, pages = {94-109}, abstract = {Weather condition are important factors affecting vehicle driving; However, the existing research results of weather factors on vehicle driving behavior are relatively scattered, and it is difficult to make effective decision analysis. Knowledge graph is one of the mainstream forms of knowledge base organization, which can clearly express the complex relationships between different objects. Therefore, introducing knowledge graph into driving behavior management is of great significance to improve decision making ability. This study analyzes the characteristics of driving behavior affected by weather, and constructs the knowledge expression model of ``weather condition–environmental condition–driving behavior''. On this basis, a representation method based on characteristic value is proposed, and then the driving behavior knowledge graph is constructed. This method can clearly express the comprehensive relationship between weather condition and driving behavior, and provide support for vehicle driving behavior decision under complex weather condition. We take the existing relevant standards in the field of Meteorology and road traffic laws and regulations as examples to verify the practicability of the method. It lays a foundation for the construction of the knowledge graph of weather factors affecting driving behavior for autonomous vehicles in the future. } @@ -107,7 +107,7 @@ @InProceedings{mlic24-12 @InProceedings{mlic24-13, title = {Light Weight Apple Defect Detection by Gaussian Mixture Model and Attention Mechanism}, - author = {Xiubo, Ma and Xiongwei, Sun and Shaoqing, Shi}, + author = {Ma, Xiubo and Sun, Xiongwei and Shi, Shaoqing}, pages = {110-119}, abstract = {To improve the speed and accuracy of online apple defect detection, a modified PP-PicoDet model based on mixed Gaussian color modeling and spatial attention mechanism is proposed. Firstly, in order to enhance the detection performance of small and low-contrast defects, the method constructs an offline color model based on GMM theory and builds a saliency numerical channel that integrates the saliency of anomalous targets on RGB data. Secondly, the SE mechanism of the network is optimized, integrating the CBAM (convolution block attention module) structure to enhance the perception of structural features in anomalous regions through strengthened spatial attention mechanisms. Furthermore, the abnormal loss function is adjusted by employing a Gaussian model to establish a label assignment loss function strategy, simplifying parameter tuning and improving the model's motivation for small proportion defects. This enhancement aims to improve the detection performance of the model under conditions of imbalanced samples. Experiments indicate that the algorithm, implemented on a self-built application platform using real-world datasets, achieved a 5.6\% increase in accuracy and a 5.8\% increase in recall at the cost of only 0.5\% time delay. The algorithm meets the requirements of timeliness and reliability for detecting defects online, contributing to the enhancement of efficiency in apple quality grading and production. } @@ -115,7 +115,7 @@ @InProceedings{mlic24-13 @InProceedings{mlic24-14, title = {Hybrid Neural Network Model for Extracting Character Relationships that Integrates Multi-level Information}, - author = {Xichao, Zhu and Zhanbin, Chen}, + author = {Zhu, Xichao and Chen, Zhanbin}, pages = {120-128}, abstract = {In the realm of natural language processing, extracting entity relationships in Chinese holds paramount importance, serving as the bedrock for various downstream tasks. However, the complexity entrenched within Chinese linguistic structures and semantic nuances presents formidable challenges. To surmount these obstacles, our study proposes a sophisticated neural network model meticulously designed to integrate multi-level information for relationship extraction. By segmenting sentences into distinct components, entity-adjacent words and the sentence. The model capitalizes on the strengths of BERT pretrained models for dynamic word embedding. Furthermore, leveraging Bidirectional Gated Recurrent Units (BiGRU) and Convolutional Neural Networks (CNN) at the sentence level enables the capture of both sequential and structural features. At the entity-adjacent word level, a fusion of two fully connected neural networks extracts intricate associations between entities and neighboring words. Rigorous ablation and comparative experiments conducted on a comprehensive corpus underscore the efficacy of our proposed approach. Remarkably, compared to benchmark methods, our model exhibits a substantial 6.61\% increase in recall rate and a note-worthy 5.31\% improvement in the F1 value. } @@ -123,7 +123,7 @@ @InProceedings{mlic24-14 @InProceedings{mlic24-15, title = {Fusing Adaptive Meta Feature Weighting for Few-shot Object Detection}, - author = {Peng, Zhou and Anzhan, Liu}, + author = {Zhou, Peng and Liu, Anzhan}, pages = {129-137}, abstract = {Few-shot object detection models often lack the perceptual ability to detect the target objects and fine-tuning the model on base class images to quickly adapt to new tasks can lead to feature shift issues. We propose an Adaptive Meta-Feature Weighting (AMFW-YOLO) object detection model for solving these problems. This model introduces an attention mechanism based on spatial and channel-wise squeeze-and-excitation (scSE) blocks, which helps the model focus on the regions of interest in the target samples and suppresses interference from background regions. Additionally, to compensate for the feature shift caused during the fine-tuning stage, we design an Adaptive Meta-Feature Weighting module (AMFW), This module embeds positional information into spatial features, captures long-range dependencies along two directions, and adaptively compensates for the weights of deep global features, effectively improving the accuracy of the model. } @@ -131,7 +131,7 @@ @InProceedings{mlic24-15 @InProceedings{mlic24-16, title = {Attention-Enhanced Pointer Network for Summarization with Key Information}, - author = {Qiming, Li and Liang, Gao}, + author = {Li, Qiming and Gao, Liang}, pages = {138-146}, abstract = {Addressing the limitations of mainstream generative text summarization models, such as poor semantic quality, inappropriate allocation of weights to key information, and constraints in extracting the semantic essence of textual content by existing natural language generation models, we propose an Attention-Augmented Pointer Generation Network (AUPT). This model utilizes TextRank technology to extract crucial information, combines positional encoding with an adaptive masking mechanism to enhance positional attention scores, emphasizing the importance of key information in the text’s semantics. Furthermore, by integrating the T5-Pegasus model with the pointer generation network, it effectively handles unknown vocabulary and replication issues, enabling more accurate and reliable semantic representations. } @@ -139,7 +139,7 @@ @InProceedings{mlic24-16 @InProceedings{mlic24-17, title = {Chinese Named Entity Recognition Method Based on Lexicon and Convolution-Integrated Self-Attention}, - author = {Wenjie, Xu and Maoting, Gao}, + author = {Xu, Wenjie and Gao, Maoting}, pages = {147-154}, abstract = {Chinese Named Entity Recognition methods primarily consist of span-based approaches and sequence-to-sequence methods. However, the former focuses solely on the recognition of entity boundaries, while the latter is susceptible to exposure bias. To address these issues, a Chinese NER method based on dictionary enhancement and self-attention fusion convolution is proposed. Initially, the text is encoded using the pre-trained model ERNIE 3.0 and lexical representations. Then, Bi-LSTM is utilized to further capture the contextual information of the sequence, resulting in the final character representations. Subsequently, a two-dimensional (2D) grid is constructed for modeling character pairs, and a feature integration layer is developed by merging self-attention mechanisms and convolution to refine and capture the interactions between characters. Finally, a joint predictor composed of a dual-affine classifier and multilayer perceptrons is used to predict entity categories. Experimental results demonstrate that this method can effectively recognize both flat and nested named entities. Compared to the current best-performing baseline models, the proposed method achieves an increase of 0.14\% and 2.53\% in F1 scores on the flat datasets Resume and Weibo, respectively, and an improvement of 0.52\% in F1 score on the nested dataset ACE2005. } @@ -147,7 +147,7 @@ @InProceedings{mlic24-17 @InProceedings{mlic24-18, title = {Conversational Recommendation System Based on Utterance Act and Emotion}, - author = {Jiahao, An and Huibo, Dang}, + author = {An, Jiahao and Dang, Huibo}, pages = {155-164}, abstract = {Conversational recommendation system (CRS) aims to acquire user preferences of conversation and then make recommendations to users. Existing CRS enhance the characterization of items by introducing external information to improve the recommendation effect, but they ignore the most essential utterance semantic attributes in the dialogue, and do not fully consider the different emotion or act feedback of users. Based on this, this paper fully explores the utterance semantics, makes the construction of user characteristics fuller by extracting the act and emotion of the utterance, and predicts the act of the conversation process, and then enhances the response through the emotional vocabulary, to improve the interaction experience between the user and the system. A large number of experiments on public datasets show that the model proposed in this paper outperforms the most advanced methods. } @@ -155,21 +155,21 @@ @InProceedings{mlic24-18 @InProceedings{mlic24-19, title = {A Sound Source Location Method Based on Time Difference of Arrival with Improved Dung Beetle Optimizer}, - author = {Chunning, Song and Jindong, Zhang}, + author = {Song, Chunning and Zhang, Jindong}, pages = {165-176}, abstract = {In microphone array sound source localization based on Time Difference of Arrival (TDOA), traditional methods for solving the nonlinear equations of TDOA lead to significant deviations and lower accuracy. To address this issue, this paper proposes a TDOA-based sound source localization method using an Improved Dung Beetle Optimizer (IDBO) algorithm. This method enhances the performance of the Dung Beetle Optimizer (DBO) by employing strategies such as chaotic mapping, golden sine, and adaptive tdistribution, and applies it to sound source localization. To evaluate the performance of the IDBO, it is compared with DBO, Harris Hawk Optimizer (HHO), Gray Wolf Optimizer (GWO), Bald Eagle Search (BES) algorithm, and Whale Optimization Algorithm (WOA). The results showed that in solving benchmark functions and localization models, it demonstrates faster convergence speed, higher localization accuracy, and better stability.} } @InProceedings{mlic24-20, title = {ChatGPT Plagiarism in the Academic Field: Exploration and Analysis of Plagiarism Effects}, - author = {Haiqiong, Luo and Hoiio, Kong}, + author = {Luo, Haiqiong and Kong, Hoiio}, pages = {177-187}, abstract = {This article explores the plagiarism problem of ChatGPT in the education field in detail and proposes a study to evaluate its plagiarism effect. First, it introduces ChatGPT as an important innovation based on GPT technology, its application and potential impact in the field of natural language processing. The article then focuses on the plagiarism concerns caused by ChatGPT in academia and points out the countermeasures some schools have taken. Then, the article raises research questions, aiming to explore the actual effect of ChatGPT in paper plagiarism detection, and describes the research methods and premise assumptions in detail. By rewriting and plagiarism checking analysis of a series of papers in the education field, the study attempts to evaluate ChatGPT's performance in terms of plagiarism and explore factors that may affect the degree of plagiarism. Finally, the article conducts a detailed statistical analysis of the experimental results and puts forward suggestions for further research to emphasize the importance of academic integrity and the call for the correct use of ChatGPT.} } @InProceedings{mlic24-21, title = {Characterization Study of Online Public Opinion Based on Natural Language Processing with Weibo Data}, - author = {Zhiyuan, An}, + author = {An, Zhiyuan}, pages = {188-197}, abstract = {In the summer of 2022, ``Ice cream assassin'' has emerged as a prominent buzzword on the Chinese Internet. With its vast user base of 582 million monthly active users, Sina Weibo serves as an ideal platform for analyzing information disseminated within its ecosystem. This platform not only enables us to discern prevailing public sentiment but also facilitates governmental efforts in shaping and regulating the public opinion landscape. This study encompasses a collection of 60,228 Weibo pertaining to ``ice cream assassin'' posted on Sina Weibo. By employing sentiment analysis algorithm based on Bert's fine-tuned model, we analyze temporal shifts in emotional trends, summarize changes in public sentiment, and categorize variations in popularity levels. Furthermore, through semantic network analysis, we identify two distinct thematic segments within this realm of public opinion. This study has important implications for assessing the impact of online public opinion on the economy and even society. } @@ -177,7 +177,7 @@ @InProceedings{mlic24-21 @InProceedings{mlic24-22, title = {Utilization of a Full Convolutional Autoencoder for the Task of Anomaly Detection in Hyperspectral Imagery}, - author = {Jingwen, Wang and Yize, Yu and Rui, Zhao and Minyi, Li}, + author = {Wang, Jingwen and Yu, Yize and Zhao, Rui and Li, Minyi}, pages = {198-205}, abstract = {The advancement of artificial intelligence has significantly improved the capability to capture back-ground features in hyperspectral images (HSI), thereby demonstrating commendable performance in the domain of hyperspectral anomaly detection (HAD). The existing approaches, however, still exhibit certain limitations: (1) The deep feature learning process lacks contextual, anomaly constraints, and prior information. (2) The priority reconstruction of the background cannot be ensured by traditional HSI anomaly detectors based on self-supervised deep learning. (3) The utilization of spatial information in hyperspectral images is limited by the fully connected deep network structure of the HSI anomaly detector. The performance of many hyperspectral anomaly detectors is limited by assumptions or presumptions regarding background and anomaly distributions, as these detectors cannot accurately account for the complex real-world distributions. The paper proposes a self-supervised full convolutional autoencoder as a solution to address these issues. The effectiveness and performance of the method were confirmed through evaluation on two real hyperspectral datasets, demonstrating superiority over nine other state-of-the-art methods. } @@ -185,7 +185,7 @@ @InProceedings{mlic24-22 @InProceedings{mlic24-23, title = {Enhancing Anomaly Detection in Medical Imaging: Blood UNet with Interpretable Insights}, - author = {Hao, Wang and Kexin, Cao}, + author = {Wang, Hao and Cao, Kexin}, pages = {206-214}, abstract = {Researching anomaly detection in medical imaging, particularly in blood samples, is crucial for enhancing diagnostic accuracy. This article is to devise a robust method using Blood UNet, a modified U Net architecture integrated with Lesion Enhancing Network (LEN) and Shape Model (SHAP), to improve anomaly detection precision. Specifically, the research involves preprocessing the BloodMNIST, training the Blood UNET model, and interpreting its predictions using LEN and SHAP. Experimental results on BloodMNIST showcase the efficacy of identifying anomalies within blood samples. This study highlights the importance of leveraging advanced techniques like LEN and SHAP in medical diagnostics, contributing to better patient care and healthcare efficiency. } @@ -193,168 +193,168 @@ @InProceedings{mlic24-23 @InProceedings{mlic24-24, title = {Research on Random Forest Regression Algorithms for Predicting Reynolds Stress Anisotropy in Separation Flow around Near-Wall Cylinder}, - author = {Kewei, Deng and Jianhong, Sun}, + author = {Deng, Kewei and Sun, Jianhong}, pages = {215-224}, abstract = {Standard RANS model is widely used in turbulence modelling and numerical simulation, due to its linear assumption of eddy viscosity, it is not suitable to accurately predict the Reynolds stress, especially in separated flows with anisotropy of Reynolds stress. In this study, a machine learning model is proposed by applying the random forest regression algorithm to learn the deviation between eddyviscosity and high-fidelity models for separation flows around nearwall cylinder. The output features which represent magnitude, shape and directions of Reynolds stress are extracted by decomposing Reynolds stress tensor, while 8 types of input features are extracted from raw local flow data sets to represent the main physical characteristics of the flow field. Both input and output features satisfy Galilean invariance, which contributes to improving prediction accuracy and generalization performance of the random forest regression model. The results show that the random forest regression model has a great potential to effectively predict the Reynolds stress anisotropy distribution of different Reynolds numbers.} } @InProceedings{mlic24-25, title = {A Tree-Structure Enhanced Transformer for Cardinality Estimation}, - author = {Mingjie, Hu and Qihang, Zhang and Jing, Ren and Hengzhu, Liu}, + author = {Hu, Mingjie and Zhang, Qihang and Ren, Jing and Liu, Hengzhu}, pages = {225-237}, abstract = {Accurate cardinality estimation is crucial for query optimization by guiding plan selection. Traditional cardinality estimation approaches often fail to provide precise estimates, leading to suboptimal query plans. In recent years, learning-based methods have emerged as a promising alternative. For tree-based learning methods, a conventional way to simply encode nearby parent-child pairs and learn by iteratively training hampers the performance. But it leads to a significant loss of structural information and incurs substantial computational overhead by the iterative training process. In this paper, we proposed a tree-structure positional encoding scheme. It can not only extract effective features for each node, but also capture the inherent structural characteristics of the tree. Based on the tree-based feature, we designed a novel transformer-based cardinality estimation model, which enhances the parallelism of the model training process and reduces the overhead caused by iterative training. On real-world datasets, our method beats the current state-of-the-art techniques, QF, by 25% in terms of mean qerror.} } @InProceedings{mlic24-26, title = {Hierarchical Clustering with Dynamic Time Warping for Automatic Detection and Labeling of Triggering Asynchrony in Patient-Ventilator Interaction during Mechanical Ventilation}, - author = {Xuan, Wang and Zunliang, Wang and Cheng, Chen and Songqiao, Liu}, + author = {Wang, Xuan and Wang, Zunliang and Chen, Cheng and Liu, Songqiao}, pages = {238-245}, abstract = {The patient-ventilator asynchronies in the ICU can significantly impact the management and prognosis of mechanical ventilation. Currently, machine learning methods have shown promise in ef- fectively identifying patient-ventilator asynchrony (PVA). However, there is still a lack of training datasets for PVA detection. In this study, a hierarchical clustering with dynamic time warping (DTW) method is presented to perform automatic identification and labeling of trigger asynchrony waveforms within our abnormal breathing cycle dataset for automatic identification and labeling of abnormal waveform datasets in mechanical ventilation, thereby reducing the workload of hand labeling. The experimental results show that our method can efficiently identify both ineffective triggering and double triggering from a large amount of abnormal ventilation data. These two types of trigger abnormalities are widely observed in patient-ventilator asynchronies (PVAs) during mechanical ventilation. Automatically identifying and annotating these abnormalities is crucial for reducing the burden of manual labeling, promoting the creation of PVA training datasets.} } @InProceedings{mlic24-27, title = {City Gas Load Forecasting Based on PCCs-CNN-LSTM Model}, - author = {Guangjun, Zai and Yuwei, Zhang and Sen, Wu and Zhao, Tian}, + author = {Zai, Guangjun and Zhang, Yuwei and Wu, Sen and Tian, Zhao}, pages = {246-252}, abstract = {The forecast of urban gas load is of great significance for the safety and stability of gas supply, to ensure the normal production activities of residents. The influence factors of sunshine duration were introduced, and the nine identified influencing factors were analyzed by Pearson correlation coefficient (PCCs). According to the correlation, the optimal input was selected one by one. The influencing factors with high correlation were used as the input of Convolutional Neural Networks (CNN) and Long short-term memory (LSTM), respectively, to forecast the daily load, monthly load and quarterly load of urban gas, and verify their accuracy and effectiveness. The results show that the optimal number of input factors for daily load forecasting and monthly load forecasting is 5, and the optimal number of input factors for quarterly load forecasting is 8. For daily load forecasting, the absolute percentage errors of monthly load forecasting and quarterly load forecasting of PCS- CNN-LSTM model are 3.94\%, 4.61\% and 5.73\% respectively. The root mean square error and mean absolute error of PCS-CNN-LSTM model are better than that of a single LSTM model.} } @InProceedings{mlic24-28, title = {Detection Method of Forest Pests Based on Attention Mechanism and Lightweight YOLOv5}, - author = {Kehao, Cha and Xudong, Song}, + author = {Cha, Kehao and Song, Xudong}, pages = {253-264}, abstract = {In view of the forestry pest identification research is less, manual identification time-consuming and labor-intensive low accuracy. An attention-based and lightweight YOLOv5 forestry pest identification method was proposed. First of all, the traditional backbone network CSPDarknet is modified to the improved ShuffleNetV2, which simplifies the network structure and makes the network more lightweight; Secondly, the hybrid attention mechanism CBAM (Convolutional Block Attention Module) is introduced to increase the perception ability of cyberspace and channel features while keeping the parameters and calculation load basically unchanged. Finally, the loss function is replaced by WIoU to improve model training to speed up model convergence and improve regression accuracy. The average detection accuracy of the improved model is 89.9\%, which is 3.6\% higher than that of the original YOLOv5s algorithm; The parameters decreased by 3213050 and the calculation amount decreased by 7.8. The improved model improves the detection accuracy and reduces the parameters and calculation amount. Compared with other advanced algorithms, the algorithm in this paper has excellent performance, which can provide reference for forest pest identification and management.} } @InProceedings{mlic24-29, title = {Hierarchical Quantization Algorithm for Deep Learning Network Models}, - author = {Xiaoqi, Mao}, + author = {Mao, Xiaoqi}, pages = {265-270}, abstract = {Deep learning network models have achieved inspiring performances in various fields such as computer vision, natural language processing, and biomedicine. However, the high computational and storage costs of the models restrain their application in resource-limited situations. However, due to the increased com-plexity and computation of deep neural networks, there are still some challenges in deploying deep learning models into real-world applications for resource-constrained devices. To address this problem, researchers have proposed various quantization algorithms to decrease the expenditure of calculation and storage in deep learning models. This thesis addresses the problem of hierarchical quantization of deep learning models and proposes a simple hierarchical quantization algorithm that aims to effectively reduce the computation and storage requirements of deep learning network models and maintain the accuracy of the models. To demonstrate the effectiveness of the proposed hierarchical quantization method, we conducted experiments on several classical deep learning models. Our experiments prove our approach can better maintain the models' accuracy while reduc-ing the storage and computation requirements compared to the traditional quantization algorithms.} } @InProceedings{mlic24-30, title = {Hybrid Classical Quantum Neural Network with High Adversarial Robustness}, - author = {Yongxi, Yang and Shibin, Zhang and Lili, Yan and Yan, Chang}, + author = {Yang, Yongxi and Zhang, Shibin and Yan, Lili and Chang, Yan}, pages = {271-279}, abstract = {As the realms of quantum computing and machine learning converge, a novel domain, termed quantum machine learning, is progressively forming within the sphere of artificial intelligence studies. Nonetheless, akin to its classical counterpart, this emerging field is not exempt from security vulnerabilities. Quantum machine learning systems, regardless of whether they process classical or quantum inputs, are susceptible to minor perturbations that can erroneously skew classification outcomes. These minute disruptions, often imperceptible to human observation, present a significant challenge in ensuring the integrity of quantum classifiers. As the complexity of quantum classifiers increases, their vulnerability also gradually grows. To mitigate this issue, this paper proposes a novel hybrid classical-quantum neural network model that enhances the model's adversarial robustness by adding a preprocessing layer for noise reduction and data reconstruction. Experiments demonstrate that this model exhibits higher efficiency and accuracy in noisy environments and against adversarial attacks.} } @InProceedings{mlic24-31, title = {Ships Collision Avoidance Based on Quadrangle Ship Domain and Reciprocal Velocity Obstacle}, - author = {Ying, Lu and Yuetao, Zhao and Yu, Shi and Likun, Wang and Miaoqi, Gu and Mingjie, Ou and Feixue, Luo and Jianhua, Xu}, + author = {Lu, Ying and Zhao, Yuetao and Shi, Yu and Wang, Likun and Gu, Miaoqi and Ou, Mingjie and Luo, Feixue and Xu, Jianhua}, pages = {280-289}, abstract = {Navigating the narrow and congested waters of the Yangtze River in China poses a significant challenge, leading to frequent ship-ship and ship-buoy collisions. In most cases of collisions between ships and buoys, the ship often hits and runs. This paper introduces a novel collision-avoidance decision method that employs the RVO (Reciprocal Velocity Obstacles) and QSD (Quantitative Ship Domain) to enable dynamic obstacle avoidance for ship-to-buoy and ship-to-ship, which complies with conventions on the international regulation for preventing collision at sea. QSD can dynamically adjust the ship domain model according to different speeds to address different encounter situations. The combination of RVO and QSD combines the dynamic ship domain with the obstacle avoidance algorithm, which makes the water transport safer than the traditional obstacle avoidance algorithm. In addition, this paper also compares the effects of VO (Velocity obstacle) and RVO, and the results indicate that RVO has smoother obstacle avoidance.} } @InProceedings{mlic24-32, title = {Brain Tumor Detection Algorithm Based on Improved YOLOv7}, - author = {Jiajun, Wu and Maoting, Gao}, + author = {Wu, Jiajun and Gao, Maoting}, pages = {290-296}, abstract = {A brain tumor detection algorithm based on improved YOLOv7 is proposed to address the problem of low image resolution in detecting small targets in brain tumor detection. By introducing the SPD module in the Backbone section, cross row convolution and pooling operations are eliminated, fine-grained feature learning is strengthened, and the accuracy of model detection is increased; Introducing CA attention mechanism to enhance the learning of more critical and effective features, further improving the efficiency and accuracy of the network model; And use the dynamic non monotonic frequency modulation loss function Wise-IoU to enhance the model's detection ability for low-quality samples. Overall improved YOLOv7 network model significantly improves the accuracy of low resolution samples and small object detection, and can be effectively applied to the detection of brain tumor images.} } @InProceedings{mlic24-33, title = {Extended F-expansion Method and Its Application to the Variable-coefficient Fractional Nonlinear Schro¨dinger Equation}, - author = {Libin, Hao and Xiaoshan, Zhao}, + author = {Hao, Libin and Zhao, Xiaoshan}, pages = {297-307}, abstract = {In this paper, under the definition of conformable fractional derivatives, we use the extended Fexpansion method and obtain the exact solution of the variable-coefficient factional nonlinear Schrodinger equation (FNLSE), including rational function solutions and Jacobi elliptic function solution. When the mode m of these solutions tends to 1 and 0, the hyperbolic function solution, triangular function solution, and light and dark solitary wave solution are obtained. The correlation diagram of the exact solution is plotted, and the effect of different parameters on the solution structure is deeply analyzed. By selecting a large number of parameters and comparing the graphical analysis of different solutions obtained using this method, we have identified properties related to the nonlinear Schro¨dinger equation with variable coefficients and summarized relevant theorems.} } @InProceedings{mlic24-34, title = {Prediction of Momentum in Tennis Using Random Forest Based on Bayesian Optimization and LSTM-ARIMA model}, - author = {Yan, Dong and Ruokai, Zhang and Yihang, Jin}, + author = {Dong, Yan and Zhang, Ruokai and Jin, Yihang}, pages = {308-315}, abstract = {Athletes' performances are often influenced by an intangible factor, momentum, which reflects the ability to perform exceptionally or consistently at a specific moment. Our model quantifies momentum and predicts match win rates, aiding athletes and coaches in optimizing their game strategies. We analyzed factors such as break points and winning streaks, employing a Random Forest Model to evaluate momentum's influence. Through the SHAP model, we established a quantifiable relationship with momentum and considered previous momentum using exponential weighted moving averages (EWMA). We developed a Gaussian Distribution Maximum Distance (GDMD) Threshold and utilized an LSTM-ARIMA model to predict momentum differences and identify turning points. The most critical factors were winning break points, running distance, and runs of success. Players are advised to be aware of their opponents' turning points and conserve energy to break them. Potential improvements include considering external factors like audience impact and expected goals, as well as incorporating more data to enhance model generalization capability.} } @InProceedings{mlic24-35, title = {Traffic Sign Detection Algorithm Based on Improved YOLOv5}, - author = {Xing, Wei and Hongqiong, Huang}, + author = {Wei, Xing and Huang, Hongqiong}, pages = {316-324}, abstract = {Due to the phenomenon of small size, complex background or high density of traffic signs, there is a certain degree of missing or false detection, which ultimately leads to the problem of reduced detection accuracy. To solve this problem, a real-time traffic sign detection algorithm based on YOLOv5s is proposed. Firstly, feature upsampling is carried out through ContentAware ReAssembly of Features upsampling operator in the neck network, which can aggregate information in the large receptive field, so that the network can get a more accurate feature map. Secondly, the normalized Gaussian Wasserstein distance is used as the similarity measure to construct the NIOU regression bounding box loss function to improve the overall performance of the model. Finally, the FasterNet module is used instead of the C3 module, which is lighter and faster. Experiments were carried out on TT100K data set. Compared with YOLOv5s, CNF-YOLO algorithm reduced Parameters by 1/5, the computing load decreased by 3GFLOPs, the detection speed increased by 18.4 frames/SEC and the weight file decreased by 2.1MB. All models were lighter. And its mAP@0.5 has also been increased by 0.5\% to enable rapid detection of traffic signs.} } @InProceedings{mlic24-36, title = {Federated Learning Algorithm based on Gaussi-an Local Differential Noise}, - author = {Fan, Wu and Maoting, Gao}, + author = {Wu, Fan and Gao, Maoting}, pages = {325-339}, abstract = {In differential privacy-based federated learning, the data of different clients are of-ten independently and identically distributed. During model training, each client's data will optimize and converge towards its own optimal direction, causing a client drift phenomenon, resulting in a decrease in accuracy and making it difficult to obtain the optimal global model. To address this issue, a federated learning al-gorithm based on local differential privacy is proposed. Each client is assigned its own control variable ci to control the model update direction, and a global control variable c is set on the server side. The SCAFFOLD algorithm is used to aggregate all client model parameters and control variables. During model training, a correction term c-ci is added when updating parameters on the client side, and the model training bias is adjusted according to the global control variable obtained from the server side in the previous round, thereby controlling the model's iterative direction towards the global optimum. Experimental results on the CIFAR-10 datasets demonstrated the effectiveness of the new algorithm.} } @InProceedings{mlic24-37, title = {Decentralized Federated Learning Algorithm Based on Federated Groups and Secure Multiparty Computation}, - author = {Fan, Wu and Maoting, Gao}, + author = {Wu, Fan and Gao, Maoting}, pages = {340-348}, abstract = {To solve the problem that the centralized federal learning based on privacy protection relies on trusted central servers, has low resistance to malicious attacks, and is prone to privacy leakage, this paper proposes a decentralized federated learning algorithm based on federated groups and secure multiparty computation. By establishing a federated group mechanism based on model relevance, each client has its own federated group, and model parameters are only transmitted among fed- erated group members, members outside the group are unable to access parameter information. Secret owners utilize secret sharing algorithms to split their model parameters into several secret shares, which are then transmitted to federated group members through secure channels. Federated group members then aggregate all transmitted secret shares by weighted averaging, and the secret owner receives the aggregated secret shares passed back from all federated group members, and then uses the secret recovery algorithms to recover secret, and obtains the updated parameter model. In the federated group, while a member becomes a Byzantine node, it is removed from the federated group, and another client is selected to join the group based on model relevance. So, each client participating in federated learning serves as both a data node and a computing node, federated learning eliminates reliance on servers and achieves decentralization. The good privacy performance of the proposed algorithm model is theoretically analyzed, and experiments on the FedML platform demonstrated that the algorithm has stronger resistance to attacks.} } @InProceedings{mlic24-38, title = {Fusion Analysis and Digital Realization of Inspection and Repair of High-Speed Railway Engineering Equipment}, - author = {Xiaoai, Zhou}, + author = {Zhou, Xiaoai}, pages = {349-365}, abstract = {High-speed railway engineering equipment is the basis of railway transportation, and transportation safety is closely related to the equipment operating state. In order to help inspectors and managers of the engineering departments to conduct inspection and repair more efficiently, solve problems such as the lack of intuitiveness of inspection status, unreasonable setting of inspection cycle days, and frequent occurrence of equipment accidents caused by seasonal weather changes, this paper uses deep learning to establish an automatic disease recognition model based on Convolutional Neural Network. Through the data collected from several high-speed railway workshops for verification, it is concluded that the model can realize the automatic recognition of disease types, the training accuracy reaches 97\%, and the verification accuracy reaches 76\%. Meanwhile, based on big data technology, this paper combines Convolutional Neural Network and Long Short-Term Neural Network, establishes the equipment status judgment model, and builds the inspection cycle algorithm based on equipment status. Through the data collected from multiple engineering departments, the equipment state judgment model can capture the key information from the inspection records and can thus accurately judge the equipment operating status. The accuracy of the predicted inspection times reaches 85.7\%. Finally, through the digital implementation case, it is fully proved that the design and fusion of these two applications can provide sufficient technical support for the inspection and repair of high-speed railway engineering equipment and can thus provide a comprehensive and reliable data basis for real-time and efficient inspection and repair decisions.} } @InProceedings{mlic24-39, title = {Research on Green Design Optimization of Ethnic Minority Architecture in Guangxi Based on Machine Learning }, - author = {Cong, Lu and Nenglang, Huang and Yang, Wu}, + author = {Lu, Cong and Huang, Nenglang and Wu, Yang}, pages = {366-372}, abstract = {Guangxi Zhuang Autonomous Region, as one of China's ethnic minority areas, possesses rich heritage resources in ethnic architecture, with dual objectives of cultural preservation and ecological sustainability. This paper explores the design principles of ethnic minority architecture in Guangxi, investigates the relationship between architecture and climate adaptability, and integrates them with digital fabrication technology. Utilizing parametric platforms and performance simulation tools, the study examines the climate adaptability of ethnic minority architecture in Guangxi. Through machine learning, models for lighting, thermal, and humidity environments specific to Guangxi's ethnic minority regions are developed. Optimization parameters for architectural design are proposed, and the reliability and accuracy of the models are demonstrated through training and testing, providing ecological design optimization strategies and references for future research on green architecture in ethnic minority areas.} } @InProceedings{mlic24-40, title = {Research on Multi-UAV Task Allocation Algorithm Considering Dynamic Priority Changes}, - author = {Yulai, He and Hua, Wu and Weizheng, Zhang}, + author = {He, Yulai and Wu, Hua and Zhang, Weizheng}, pages = {373-382}, abstract = {In recent years, unmanned aerial vehicles (UAVs) have found increased application across various domains. Clusters of Multi-UAV can accomplish more complex tasks, optimizing overall efficiency through rational task allocation. In practical scenarios, they exhibit distinct advantages and characteristics. However, the allocation of tasks to Multi-UAV in special environments or emergency conditions poses a widely studied problem. Existing research or methods often impose fixed task priorities (task sequences) during Multi-UAV task allocation. Yet, real-world UAV operations may witness fluctuations in task priorities due to environmental changes or human factors. For instance, areas experiencing a drop in temperature may heighten the urgency of certain supplies, or sudden outbreaks of disease may increase the demand for medical supplies. In such cases, conventional planning methods become inadequate. Hence, this paper addresses these scenarios by proposing a model for dynamic task priority changes in Multi-UAV task allocation within special environments. This thesis introduces an improved genetic algorithm, termed the improved partheno geneticgreedy combination algorithm. Through comparative experiments, the efficacy of the proposed algorithm in addressing dynamic priority changes in Multi-UAV collaborative task allocation problems is validated, enhancing problem-solving efficiency.} } @InProceedings{mlic24-41, title = {Research on Imbalanced Classification Problem Based on Optimal Random Forest Algorithm}, - author = {Shan, Yue and Hui, Liu and Zheng, He and Qiaoling, Yong and Yali, Wang}, + author = {Yue, Shan and Liu, Hui and He, Zheng and Yong, Qiaoling and Wang, Yali}, pages = {383-392}, abstract = {In order to solve the binary classification problem of imbalanced data, an optimal random forest algorithm GWORF (Grey Wolf Optimizer Random Forest) is pro-posed. The algorithm first uses BLSMOTE (BorderLine SMOTE) technology to oversample the imbalanced data set to make the positive and negative data equivalent, and then uses the Grey Wolf optimization algorithm to calculate the optimal parameters, and then puts the calculated optimal parameters into the forest for modeling training. Through testing on four imbalanced data sets, the effectiveness of the GWORF algorithm in the study of imbalanced binary classification problems is verified.} } @InProceedings{mlic24-42, title = {Forecasting Tennis Player Matches Based on Machine Learning}, - author = {Rui, Bai and Kar Hing, Chong and Haoyuan, Li and Jia Yew, Teh}, + author = {Bai, Rui and Chong, Kar Hing and Li, Haoyuan and Teh, Jia Yew}, pages = {393-403}, abstract = {This paper aims to highlight the extensive potential of analytics with the use of machine learning to improve sports modelling. We propose a supervised machine learning approach to further extend the optimization of machine learning in predicting the flow of points in tennis matches. Using data sourced from the 2023 Wimbledon Gentlemen's singles matches, we used Grey Relational Analysis and membership functions from fuzzy set theory to extract and rank 7 features that exhibit impactful ties with the player's match outcome, which includes player's serve status, games won in current set, ranking difference, distance covered, serve speed, previous victory status, and unforced error, respectively. We implemented these features to build 3 supervised models and compare their predictive performances, namely K-Nearest Neighbours, XGBoost and Logistic Regression. We adopted a train test split measure of 300 training sets and 100 testing sets. Using performance metrics such as confusion matrices, ROC curves, F1, Precision, Recall, and Accuracy scores, we constructed a scoring table to rank implemented models. Our results demonstrated that XGBoost exhibited the most significant predictive performance, followed by KNN and Logistic Regression. 5-Fold cross-validation feature stability and sensitivity analysis suggests that the feature space cre- ated is robust and stable where features are not easily subject to change in short-term predictions.} } @InProceedings{mlic24-43, title = {Research on Personalized Music Recommendation Model Based on Personal Emotion and Collaborative Filtering Algorithm}, - author = {Yingqiang, Wang and Elcid, A.Serrano}, + author = {Wang, Yingqiang and A.Serrano, Elcid}, pages = {404-412}, abstract = {In today’s digital age, music plays an important role in people's lives, but the current music recommendation system is mainly based on content, the use of collaborative filtering and other algorithms, can not according to the user's real-time emotional state, recommend suitable for the current mood of the music. This paper aims to design and implement a personalized music recommendation model based on personal emotional information and collaborative filtering algorithm. The model mainly includes two sub-models: the emotion and music selection tendency model and the music recommendation model based on collaborative filtering algorithm. By analyzing the user's emotional state and music preference, the model provides music recommendation services that are more in line with the user's psychological state, so as to improve the user experience and recommendation accuracy. The model designed in this paper can effectively solve the problem that personalized music recommendation cannot be performed according to the user's real-time emotion. At the same time, it can also be used as a reference for other personalized recommendation models.} } @InProceedings{mlic24-44, title = {Reordering CAE Matrix using Hierarchical Clustering}, - author = {Shanhong, He and Yahui, Yang and Lihong, Tang and Tao, Zhang and Xiaolong, Jiang}, + author = {He, Shanhong and Yang, Yahui and Tang, Lihong and Zhang, Tao and Jiang, Xiaolong}, pages = {413-420}, abstract = {Traversing a high-dimensional mesh using 1-D trajectory is a widely used technique in the fields of CAE computing and data management. Such a trajectory is widely known as a space-filling curve. Nevertheless, most of the space filling curves, such as Z-curve or Hilbert curve, are designed for structured meshes. Therefore, it is vital to design an effective space-filling curve for unstructured meshes. In this paper, we propose a space-filling curve for unstructured mesh by considering the original problem as a graph clustering problem. We generate a hierarchical clustering schema and uses depth-first search to traverse the hierarchical schema. In this way, the sequence of the depth-first search naturally be-comes a 1-D trajectory. Compared with traditional space-filling curves, our solution can effectively handle traversing problems on unstructured meshes.} } @InProceedings{mlic24-45, title = {Polynomial Fitting Based on Integrable Deep Neural Networks for Landau-Energy of Ferroelectrics}, - author = {Wenyu, Zhang and Yabin, Yan}, + author = {Zhang, Wenyu and Yan, Yabin}, pages = {421-429}, abstract = {Fitting the Landau-Energy polynomial has always been challenging because it is difficult to directly obtain Landau-Energy data for coefficient fitting. One possible approach to address this problem is to handle the derivative of the Landau-Energy polynomial with respect to the second-order polar- ization (dielectric constant) to obtain relevant information about the Landau-Energy. This chapter will introduce a method based on integrable neural networks to obtain an approximate model for the Landau-Energy polynomial and its parameters.} } @InProceedings{mlic24-46, title = {A Multi-attribute Large Group Decision-making Method Based on Interval-valued Pythagorean Fuzzy Number}, - author = {Xinshang, You and Zhe, Zhou}, + author = {You, Xinshang and Zhou, Zhe}, pages = {430-442}, abstract = {Firstly, a large group clustering algorithm based on data similarity is proposed, which can set different thresholds to cluster the decision results of expert groups. Secondly, the interval-valued pythagorean fuzzy number (IVPFN) hesitancy accuracy function and hesitancy score function are proposed. And based on the consideration of the distance between centroids and rectangular area in the geometric meaning of IVPFN, a calculation formula is proposed to distinguish different IVPFNs. Thirdly, use the above formula to construct a weight calculation model for evaluation criteria with adjustment coefficients. The decision matrix is weighted and its relative distance from the positive and negative ideal solutions is calculated to produce the final ranking. Finally, the cultural tourism project decision-making problem is analyzed as an arithmetic example and compared with the methods in the literature of other related fields to illustrate the rationality and scientificity of this paper.} } @InProceedings{mlic24-47, title = {Construction and Evaluation of a Metabolic Syndrome Prediction Model Based on Classification Algorithms}, - author = {Zhenghao, Zhao and Qiongqiong, Hu}, + author = {Zhao, Zhenghao and Hu, Qiongqiong}, pages = {443-455}, abstract = {Exploring the risk factors influencing Metabolic Syndrome (MS), constructing a risk prediction model based on multiple classification algorithms, comparing the predictive performance of different models for MS, and interpreting the models to derive specific MS prediction rules, providing scientific basis for MS prediction. A retrospective analysis was conducted on clinical data from 2,193 MS patients. Based on whether the patients developed MS, they were divided into a healthy group and an MS group. Statistical correlation tests were used to identify the risk factors associated with MS. Six classification algorithms, including decision trees, logistic regression, random forests, naive Bayes, K-nearest neighbors, and support vector machines, were employed to build an MS prediction model. The prediction model's performance was evaluated using R language by generating receiver operating characteristic (ROC) curves. Among the 2,193 MS patients, the incidence rate of MS was 34.66\%. Significant differences (P¡0.05) were observed between the healthy group and the MS group in terms of age, marital status, income, ethnicity, waist circumference, body mass index (BMI), uric acid levels, blood glucose levels, high-density lipoprotein levels, and triglyceride levels. Blood glucose, waist circumference, BMI, and triglycerides showed a significant linear correlation with MS. The ROC curve results indicated that the random forest algorithm achieved an area under the curve (AUC) of 0.94 (95\% CI: 0.914-0.957), logistic regression achieved an AUC of 0.90 (95\% CI: 0.867-0.925), support vector machines achieved an AUC of 0.89 (95\% CI: 0.859-0.920), decision trees achieved an AUC of 0.87 (95\% CI: 0.831-0.905), K-nearest neighbors achieved an AUC of 0.81 (95\% CI: 0.770-0.850), and naive Bayes achieved an AUC of 0.74 (95\% CI: 0.694-0.785). The study results confirmed that factors such as age, marital status, waist circumference, BMI, blood glucose levels, and triglyceride levels are all risk factors for developing MS. Furthermore, the random forest and logistic regression models demonstrated excellent performance in predicting MS.} }