Academic Journal of Science and Technology ISSN: 2771-3032 | Vol. 10, No. 1, 2024 305 Rumor Detection with A Novel Graph Neural Network Approach Tianrui Liu1, *, Qi Cai2, Changxin Xu3, Bo Hong4, Fanghao Ni5, Yuxin Qiao6, Tsungwei Yang7 1 Electrical and Computer Engineering, University of California San Diego, La Jolla, USA 2 Computer Science and Engineering, University of North Texas, Denton, USA 3 Computer Information Technology, Northern Arizona University, Flagstaff, USA 4 Computer Information Technology, Northern Arizona University, Flagstaff, USA 5 Data Science and Artificial Intelligence, Indian Institute of Technology Guwahati, Assam, India 6 Computer Information Technology, Northern Arizona University, Flagstaff, USA 7 Computer Science, Tunghai University, Taichung, Taiwan * Corresponding author: Tianrui Liu (Email: tianrui.liu.ml@gmail.com) Abstract: The wide spread of rumors[3] on social media has caused a negative impact on people's daily life, leading to potential panic, fear and mental health problems for the public.[47] How to debunk rumors as early as possible remains a challenging problem. Existing studies mainly leverage information propagation structure to detect rumors[6], while very few works focus on correlation among users that they may coordinate to spread rumors in order to gain a large popularity. In this paper, we propose a new detection model, that jointly learns both the representations of user correlation and information propagation to detect rumors on social media. Specifically, we leverage graph neural networks to learn the representations of user correlation from a bipartite graph[5] that describes the correlations between users and source tweets[12], and the representations of information propagation with a tree structure. Then we combine the learned representations from these two modules to classify the rumors. Since malicious users intend to subvert our model after deployment, we further develop a greedy attack scheme to analyze the cost of three adversarial attacks: graph attack, comment attack and joint attack. Evaluation results on two public datasets illustrate that the proposed MODEL outperforms the state-of-the-art rumor detection models. We also demonstrate our method performs well for early rumor detection.[8] Moreover, the proposed detection method is more robust to adversarial attacks compared to the best existing method. Importantly, we show that it requires high cost for attackers to subvert user correlation pattern, demonstrating the importance of considering user correlation for rumor detection. Keywords: Rumor detection, Natural language processing, Graph neural networks. 1. Introduction This paper proposes a novel rumor detection model, which jointly learns the representations of user correlation and information propagation to detect rumors on social media. Nowadays, rumors are like social virus that spread fast, deep and broadly across social media, such as Twitter and Facebook, which has caused potential threat and panic to people's daily life. For example, a large number of rumors about COVID-19 pandemic make a ton of people feel uneasy and scared in the year of 2020. Hence, how to effectively detect rumors[2] and prevent them from diffusion as early as possible has attracted a lot of attention from both industry and academia in recent years. The early works mainly adopt traditional machine learning methods[59] based on handcrafted features, such as user profile and text contents, to detect rumors online. However, the traditional approaches, such as SVM, Decision Tree and Random Forest, heavily depend on feature engineering[53], which is time- consuming and can be easily manipulated by malicious users. Recent studies developed deep learning models, such as bidirectional transformers(BERT)[10](which are widely using in review tasks[13]), recurrent neural network(RNN),[27] recursive neural networks (RvNN)[39], Long-Short Term Memory (LSTM), generative adversarial network (GAN)[16,56], transformer[7,20] and Convolutional Neural Networks (CNN)[11,17], to learn sequential features from information propagation patterns over time.[1] These models also have widespread applications in other fields, such as data security [21,22], vision learning [18,52], material analysis [15,51] , compiling [33] and hardware designing[34], E-commerce[37], image segmentation[43], traffic controlling [38], communication [28,32] and Aerial Search [30] . These methods, however, only learn the correlations from local neighbors in the structure of information propagation while ignore the global structures of rumor dispersion.[32] In order to better learn the representations of misinformation propagation, researchers recently developed a graph convolutional neural networks (GCN)[40] based detection model, BiGCN, that combines top-down with bottom-up tree propagation structure to deal with the propagation and dispersion of rumors. While BiGCN improves the detection accuracy, it does not take user behaviors of misinformation campaigns into account. In practice, some malicious users coordinate to report some same claims together in order to make rumors spread faster and more broadly within a short time across social media. This global user dependency relationship is helpful to find the correlations among different users and source tweets. It thus can improve the detection accuracy of the coordinated misinformation campaigns.[14] Inspired by this insight, we propose a joint detection model, called MODEL, which encodes both user correlation pattern and information propagation pattern to improve the accuracy of rumor detection. The proposed method learns the representations from a user-tweet bipartite graph and from a tree structure of information diffusion, as shown in Fig. The user-tweet bipartite graph, consisting of users and source 306 tweets, denotes users perform actions, such as post, retweet and comment, on the source tweets. [46]The tree structure of information propagation shows the information of a source tweet forwards from a root node (source tweet) to its children’s nodes (comments/retweet) in a tree.[29] After that, we combine the learned representations from these two parts and then feed them into a fully connected layer[55] to classify the source tweets into different categories. In order to enhance the influence of root tweets in a deep tree, we use the same methodology as BiGCN to pass the features of root tweets to their children’s nodes.[44] The main contributions of this paper are summarized as follows: We propose a new detection model, which learns the joint representations of user correlation and information propagation to detect rumors on social media. We adopt a greedy attack scheme that generates fake edges and comments to explore the cost of adversarial attacks on the proposed method. Evaluations on two public datasets demonstrate the proposed model outperforms the baselines in terms of detection accuracy and F1 score. We also show attackers need higher cost to fool our method compared to the best baseline. The rest of paper is organized as follows. Section 1 introduces the problem of rumor detection and the background of graph neural networks. Then we present the overall architecture of the proposed model in Section 2 and 3 In Section 4, we evaluate the performance of the proposed model on two public datasets. Finally, we conclude the paper in Section 5. 2. Organization of the Text We first introduce the problem of rumor detection, and then present the background of graph convolutional neural networks used in the proposed model. 2.1. Problem Statement Figure 1. A toy example of rumor detection Fig.1 illustrates an example of our rumor detection method, consisting of two parts: user correlation and information propagation. For user correlation, let 𝑇 , 𝑇 , … 𝑇 denote a set of 𝑛 source claims (tweets) made by many different users on social media, 𝑈 , 𝑈 , … 𝑈 denote a group of 𝑘 users who perform actions, such as post and retweet, on source tweets. In this paper, a bipartite graph, 𝐺 𝑉, 𝐸, 𝑨 is constructed to describe the user-tweet relationship, where 𝑉 denotes the vertices, and 𝐸denotes the edges. When a user 𝑈 performs an action on a source tweet 𝑇 , there is an edge between them. Mathematically, the relationship between users and source tweets can be denoted by an adjacency matrix. For information propagation module, each source tweet 𝑇 consists of 𝑚 comments with a tree structure of information diffusion, denoted by 𝐶 , 𝐶 , ⋯ , 𝐶 . We use $𝐺 𝑉, 𝐸, 𝑨 to denote the tree structure. Similar to user-tweet relationship, we adopt an adjacency matrix, 𝑨 , to denote the relationship among (root) source tweets and the corresponding comments in a tree structure. The goal of this paper is to learn the representation of source tweets from both user correlation and information propagation, and then classify the source tweets into 𝑑 different categories based on the learned representations. 2.2. Graph Convolutional Neural Network Graph neural networks (GNNs) aim to encode[23] structural information about a graph into a low-dimensional latent space, which generally consists of two components 1) graph convolution layers which extract substructure features from local nodes; 2) a graph aggregation layer that aggregates node-level features into a graph-level representation. As one of the most popular GNN, graph convolutional neural networks (GCNs) have been widely used in a variety of applications[36], such as classification and clustering,[54] because they can encode both node features and graph structure of neighboring nodes. The basic idea of GCNs is to learn the representations of nodes in a graph based on the input features of nodes, 𝑿, and their graph structure, denoted by an adjacency matrix 𝑨. While there are many different variants of GCNs in the literature, the general form of a GCN is expressed by: 𝑯𝒍 𝟏 𝜎 𝑨𝑯 𝑾 , Where 𝐴 𝐷 𝐴𝐷 is the normalized adjacency matrix. In addition, 𝑊 is the weight matrix to be trained, and 𝜎 ⋅ is the activation function, such as ReLU.[41] 3. System Design and Modeling In this section, we introduce the overall architecture of the proposed our model for rumor detection. The core idea of our model is to learn the representations of source tweets from both user correlation and information propagation respectively, which are then adopted to classify rumors. The reason why we learn these two representations independently is that the embeddings of user features and comments may be not in the same latent space. The components of the model are discussed below: 3.1. User feature encoding To reduce account maintenance cost, many user accounts would be created within a short period of time. Inspired by this, we attempt to utilize user profiles as features to depict how users participate in rumor dispersion.[4] The extracted features include number of followers, number of friends, account created time, the total number of tweets and so on. We feed user features into fully-connected layers to learn its embedding for further optimization[45] 3.2. Text encoding In order to capture the sequential information present in source tweets and the corresponding responses such as comments, this paper employs the Gated Recurrent Unit (GRU) as an encoder to derive a feature vector.[9] Given the 307 variance in sequence length of tweets, zero padding is performed to standardize the length of each sequence. The text data from tweets is then embedded into a lower dimension,[50] and pre-trained word embeddings specific to Twitter data, provided by the GloVe algorithm, are utilized. [24] These embeddings are then fed into the GRU to learn the representation of the Twitter text content. This approach facilitates a nuanced understanding of the sequential nature of the text data, which is instrumental in analyzing and detecting rumors on social media platforms.[25] Figure 2. Our proposed model architecture 3.3. User Correlation Representation In addressing the scenario where a collective of users may collaborate to propagate false claims swiftly and extensively across social media, this paper delves into exploring correlations among different users and their claims. To achieve this, a user correlation pattern is utilized to learn the representations of source tweets within a user-tweet graph. A bipartite graph is constructed to delineate the relationship between users and tweets, which then serves as a foundation for applying a two-layer Graph Convolutional Neural Networks (GCN) to learn the representations of source tweets. To facilitate the scalability of the algorithm, especially in the context[19] of large-scale graphs encompassing hundreds of thousands of nodes, a subgraph approach is adopted. By focusing on a subgraph with k-hop neighboring nodes of source tweets as input for the GCN model, the training time for graph-based rumor detection is significantly reduced[57], making the methodology more efficient and applicable to large-scale scenarios.[35] This streamlined approach aims to enhance the practicality and effectiveness of the rumor detection process, especially in a large-scale, real-world social media environment.[31] 3.4. Information Propagation Representation In order to understand the flow of source tweets to their respective child nodes, a deep learning model rooted in a tree propagation structure is devised to learn their representations. The methodology employed draws parallels to a referenced approach for constructing a top-down tree concerning a source tweet and its ensuing comments. The merit of utilizing a top-down tree lies in its capability to aptly capture the causal features intrinsic to a rumor propagation thread. Analogous to the approach taken for user correlation representation, a Graph Convolutional Neural Networks (GCN) with a directional graph is adopted to learn the representations of information propagation. This method is geared towards effectively understanding the dynamics of rumor spread and the interactions within a thread, which is crucial for accurate rumor detection.[42] 3.5. Rumor Classification with Joint Representations The methodology employs the combined representations of user correlation and information propagation to facilitate rumor detection. Initially, the representations concerning user correlation and information propagation are concatenated. Following this step, the merged representations are fed into fully connected layers, culminating in a softmax layer.[49] This structure aids in categorizing the rumors into distinct classifications. By harnessing both the interactions among users and the dissemination patterns of information, this procedure aspires to enhance the accuracy of rumor detection across social media platforms.[60] 3.6. Model training The proposed model is trained by minimizing the Cross- Entropy loss between the predictions and the ground truth. During training, the model parameters are updated using back propagation until the loss converges or the maximum iterations are met.[48] To summarize, in this system, we first compute the representations of source tweets from both user correlation and information propagation, and then feed them into fully connected layers to classify the source tweets. 4. Evaluation In this section, extensive experiments are carried out to evaluate the performance of the proposed model on two benchmark datasets, Twitter15 and Twitter 16. First of all, we compare the detection accuracy of our method with the baseline methods. In order to demonstrate the robustness of our method, we further analyze the cost of adversarial attack. Also, we demonstrate the good performance of the proposed method for early rumor detection.[58] Finally, we conduct ablation studies to compare MODEL with its variants. 4.1. Dataset To evaluate the performance of the proposed MODEL, we conduct extensive experiments on two public Twitter datasets, Twitter15 and Twitter16. These two datasets contain four labels: Non-rumor (N), False Rumor (F), True Rumor (T), and Unverified Rumor (U). To construct a completely connected user-tweet graph, we choose users that report at least two source tweets or responsive post. Since user profiles are not available in the original datasets, we collect them from Twitter API 1. Table 1 illustrates the statistics of the two benchmark datasets.[61] Note that, the total number of source tweets in this paper is less than that in the original datasets, 308 because some source tweets and users are suspended at the time of collecting data[26] from Twitter this year. Like the previous studies, we split the datasets into training data and testing data with 80% and 20%, respectively. In addition, to better evaluate the performance of our method, we adopt 𝑘- fold (𝑘 = 5) cross-validation method to conduct experiments in this paper. Table 1. Twitter datasets statistics Statistics Twitter15 Twitter16 # of Source 1473 790 # of User applied 47925 24047 # of True rumors 366 203 # of False rumors 360 192 # of Unverified 372 196 # of Non-rumors 374 199 Table 2. Experiment results 4.2. Experiment results We first compare the detection performance of the proposed MODEL with baselines above on Twitter15 dataset. We use accuracy and F1 score as the evaluation metrics in the experiments. Table 2 illustrates the comparison results of different rumor detection models. It can be observed that our method outperforms the baseline methods in term of accuracy and F1 score. The reason why our method beats BiGCN is that we can learn the correlations among different sources tweets with user correlation pattern while the latter does not. Thus, we can see the effectiveness of incorporating user correlation pattern for rumor detection. For Twitter16 dataset, we also show that the proposed method has a better detection performance than the baselines, as shown in Table 2. 5. Conclusion Rumor detection algorithms are essential for identifying and curbing the spread of misinformation online. They ensure that only verified information circulates, safeguarding public safety, protecting financial markets, and maintaining national security. Advanced models, especially those using graph neural networks, enhance early detection and accuracy, mitigating the potential negative impacts of false information swiftly. In this paper, we developed a novel rumor detection model, that jointly learns the representations of user correlation and information propagation based on graph neural networks. Specifically, it first leverages graph convolutional neural networks (GCNs) to learn the representations of source tweets from a user-tweet bipartite graph and a tree propagation structure, respectively. Then we combine the learned representations to classify rumors. Evaluations on two publicly available datasets show that our model has the best detection performance compared to the existing detection models. It also has a higher detection accuracy for early rumor detection compared to the baselines. In the future, we would like to investigate more powerful approaches to learn representation of text information, which is a viable way to gain a better model performance. References [1] Atwood, James, and Don Towsley. "Diffusion-convolutional neural networks." Advances in neural information processing systems 29 (2016). [2] Bhutani, Bhavika, et al. "Fake news detection using sentiment analysis." 2019 twelfth international conference on contemporary computing (IC3). IEEE, 2019. [3] Bian, Tian, et al. "Rumor detection on social media with bi- directional graph convolutional networks." Proceedings of the AAAI conference on artificial intelligence. Vol. 34. No. 01. 2020. [4] Liu, Tianrui, et al. "News Recommendation with Attention Mechanism." Journal of Industrial Engineering and Applied Science 2.1 (2024): 21-26. [5] Hamilton, Will, Zhitao Ying, and Jure Leskovec. "Inductive representation learning on large graphs." Advances in neural information processing systems 30 (2017). [6] Choi, Daejin, et al. "Rumor propagation is amplified by echo chambers in social media." Scientific reports 10.1 (2020): 310. [7] Li, Yanjie, et al. "Transfer-learning-based network traffic automatic generation framework." 2021 6th International Conference on Intelligent Computing and Signal Processing (ICSP). IEEE, 2021. [8] Ma, Jing, Wei Gao, and Kam-Fai Wong. "Rumor detection on twitter with tree-structured recursive neural networks." Association for Computational Linguistics, 2018. [9] Pennington, Jeffrey, Richard Socher, and Christopher D. Manning. "Glove: Global vectors for word representation." Proceedings of the 2014 conference on empirical methods in natural language processing (EMNLP). 2014. [10] Devlin, Jacob, et al. "Bert: Pre-training of deep bidirectional transformers for language understanding." arXiv preprint arXiv:1810.04805 (2018). [11] Zhao, Yufan, et al. "AN EXAMINATION OF TRANSFORMER: PROGRESS AND APPLICATION IN THE FIELD OF COMPUTER VISION." СОВРЕМЕННАЯ 309 НАУКА: АКТУАЛЬНЫЕ ВОПРОСЫ, ДОСТИЖЕНИЯ И ИННОВАЦИИ. 2023.s [12] Xiong, Jize, et al. "Decoding sentiments: Enhancing covid-19 tweet analysis through bert-rcnn fusion." Journal of Theory and Practice of Engineering Science 4.01 (2024): 86-93. [13] Zhao, Zhiming, et al. "Enhancing E-commerce Recommendations: Unveiling Insights from Customer Reviews with BERTFusionDNN." Journal of Theory and Practice of Engineering Science 4.02 (2024): 38-44. [14] Fu, Zhe, Xi Niu, and Li Yu. "Wisdom of crowds and fine- grained learning for serendipity recommendations." Proceedings of the 46th International ACM SIGIR Conference on Research and Development in Information Retrieval. 2023. [15] Wei, Xiaojun, et al. "Narrowing Signal Distribution by Adamantane Derivatization for Amino Acid Identification Using an α-Hemolysin Nanopore." Nano Letters (2024). [16] Song, Ge, et al. "Energy consumption auditing based on a generative adversarial network for anomaly detection of robotic manipulators." Future Generation Computer Systems 149 (2023): 376-389. [17] Ou, Junlin, et al. "Hybrid path planning based on adaptive visibility graph initialization and edge computing for mobile robots." Engineering Applications of Artificial Intelligence 126 (2023): 107110. [18] Zhou, Yucheng, et al. "Visual In-Context Learning for Large Vision-Language Models." arXiv preprint arXiv: 2402.11574 (2024). [19] Zhou, Yucheng, et al. "Thread of thought unraveling chaotic contexts." arXiv preprint arXiv:2311.08734 (2023). [20] Zhou, Yucheng, et al. "Claret: Pre-training a correlation-aware context-to-event transformer for event-centric generation and classification." arXiv preprint arXiv:2203.02225 (2022). [21] Tian, Jiwei, et al. "LESSON: Multi-Label Adversarial False Data Injection Attack for Deep Learning Locational Detection." IEEE Transactions on Dependable and Secure Computing (2024). [22] Tian, Jiwei, et al. "Adversarial attacks and defenses for deep- learning-based unmanned aerial vehicles." IEEE Internet of Things Journal 9.22 (2021): 22399-22409. [23] Wu, Jing, et al. "SwitchTab: Switched Autoencoders Are Effective Tabular Learners." arXiv preprint arXiv :2401.02013 (2024). [24] Chen, Suiyao, et al. "Recontab: Regularized contrastive representation learning for tabular data." arXiv preprint arXiv:2310.18541 (2023). [25] Chen, Suiyao, et al. "Claims data-driven modeling of hospital time-to-readmission risk with latent heterogeneity." Health care management science 22 (2019): 156-179. [26] Chen, Suiyao, et al. "A data heterogeneity modeling and quantification approach for field pre-assessment of chloride- induced corrosion in aging infrastructures." Reliability Engineering & System Safety 171 (2018): 123-135. [27] Hsieh, Yung-Ting, Khizar Anjum, and Dario Pompili. "Ultra- low Power Analog Recurrent Neural Network Design Approximation for Wireless Health Monitoring." 2022 IEEE 19th International Conference on Mobile Ad Hoc and Smart Systems (MASS). IEEE, 2022. [28] Hsieh, Yung-Ting, Zhuoran Qi, and Dario Pompili. "ML-based joint Doppler estimation and compensation in underwater acoustic communications." Proceedings of the 16th International Conference on Underwater Networks & Systems. 2022. [29] Sun, Chuanneng, et al. "Fed2kd: Heterogeneous federated learning for pandemic risk assessment via two-way knowledge distillation." 2022 17th Wireless On-Demand Network Systems and Services Conference (WONS). IEEE, 2022. [30] Sun, Chuanneng, Songjun Huang, and Dario Pompili. "HMAAC: Hierarchical Multi-Agent Actor-Critic for Aerial Search with Explicit Coordination Modeling." 2023 IEEE International Conference on Robotics and Automation (ICRA). IEEE, 2023. [31] Sun, Chuanneng, et al. "Contextual Biasing of Named-Entities with Large Language Models." arXiv preprint arXiv: 2309.00723 (2023). [32] Younis, Ayman, Chuanneng Sun, and Dario Pompili. "Communication-efficient Federated Learning Design with Fronthaul Awareness in NG-RANs." 2022 IEEE 19th International Conference on Mobile Ad Hoc and Smart Systems (MASS). IEEE, 2022. [33] Flynn, Patrick, Xinyao Yi, and Yonghong Yan. "Exploring source-to-source compiler transformation of OpenMP SIMD constructs for Intel AVX and Arm SVE vector architectures." Proceedings of the Thirteenth International Workshop on Programming Models and Applications for Multicores and Manycores. 2022. [34] Yi, Xinyao, et al. "CUDAMicroBench: Microbenchmarks to Assist CUDA Performance Programming." 2021 IEEE International Parallel and Distributed Processing Symposium Workshops (IPDPSW). IEEE, 2021. [35] Li, Xiaying, Belle Li, and Su-Je Cho. "Empowering Chinese Language Learners from Low-Income Families to Improve Their Chinese Writing with ChatGPT’s Assistance Afterschool." Languages 8.4 (2023): 238. [36] Xie, Ying, et al. "Advancing Legal Citation Text Classification A Conv1D-Based Approach for Multi-Class Classification." Journal of Theory and Practice of Engineering Science 4.02 (2024): 15-22. [37] Luo, Yang, et al. "Enhancing E-commerce Chatbots with Falcon-7B and 16-bit Full Quantization." Journal of Theory and Practice of Engineering Science 4.02 (2024): 52-57. [38] Pan, Zhenyu, et al. "Ising-traffic: Using ising machine learning to predict traffic congestion under uncertainty." Proceedings of the AAAI Conference on Artificial Intelligence. Vol. 37. No. 8. 2023. [39] Pan, Zhenyu, et al. "CoRMF: Criticality-Ordered Recurrent Mean Field Ising Solver." arXiv preprint arXiv: 2403.03391 (2024). [40] Hu, Zhirui, et al. "On the design of quantum graph convolutional neural network in the nisq-era and beyond." 2022 IEEE 40th International Conference on Computer Design (ICCD). IEEE, 2022. [41] Chen, Yinda, et al. "Self-supervised neuron segmentation with multi-agent reinforcement learning." arXiv preprint arXiv: 2310.04148 (2023). [42] Chen, Yinda, et al. "Learning multiscale consistency for self- supervised electron microscopy instance segmentation." arXiv preprint arXiv:2308.09917 (2023). [43] Chen, Yinda, et al. "Generative text-guided 3d vision-language pretraining for unified medical image segmentation." arXiv preprint arXiv:2306.04811 (2023). [44] Tong, Xin, et al. "A Deep‐Learning Approach for Low‐Spatial‐ Coherence Imaging in Computer‐Generated Holography." Advanced Photonics Research 4.1 (2023): 2200264. [45] Xu, Renjun, et al. "$ E (2) $-Equivariant Vision Transformer." Uncertainty in Artificial Intelligence. PMLR, 2023. 310 [46] Gao, Shangde, et al. "Contrastive Knowledge Amalgamation for Unsupervised Image Classification." International Conference on Artificial Neural Networks. Cham: Springer Nature Switzerland, 2023. [47] Shangguan, Zhongkai, Zihe Zheng, and Lei Lin. "Trend and thoughts: Understanding climate change concern using machine learning and social media data." arXiv preprint arXiv:2111.14929 (2021). [48] Shangguan, Zhongkai, et al. "Neural process for black-box model optimization under bayesian framework." arXiv preprint arXiv:2104.02487 (2021). [49] Zang, Hengyi. "Precision calibration of industrial 3d scanners: An ai-enhanced approach for improved measurement accuracy." Global Academic Frontiers 2.1 (2024): 27-37. [50] Wang, Lingyun, Jose A. Sahel, and Shaohua Pi. "Sub2Full: split spectrum to boost OCT despeckling without clean data." arXiv preprint arXiv:2401.10128 (2024). [51] Wang, Lingyun, et al. "Axial super-resolution optical coherence tomography via complex-valued network." Physics in Medicine & Biology 68.23 (2023): 235016. [52] Wang, Yishuang, et al. "Predicting Nonlinear Structural Dynamic Response of ODE Systems Using Constrained Gaussian Process Regression." Society for Experimental Mechanics Annual Conference and Exposition. Cham: Springer Nature Switzerland, 2023. [53] Platz, Roland, Xinyue Xu, and Sez Atamturktur. "Introducing a Round-Robin Challenge to Quantify Model Form Uncertainty in Passive and Active Vibration Isolation." Society for Experimental Mechanics Annual Conference and Exposition. Cham: Springer Nature Switzerland, 2023. [54] Li, Lei. "CPSeg: Finer-grained Image Semantic Segmentation via Chain-of-Thought Language Prompting." Proceedings of the IEEE/CVF Winter Conference on Applications of Computer Vision. 2024. [55] Li, Lei. "Segment Any Building." Computer Graphics International Conference. Cham: Springer Nature Switzerland, 2023. [56] Li, Lei, et al. "Mask-FPAN: semi-supervised face parsing in the wild with de-occlusion and UV GAN." Computers & Graphics 116 (2023): 185-193. [57] Ni, Jianyuan, et al. "Physical-aware Cross-modal Adversarial Network for Wearable Sensor-based Human Action Recognition." arXiv preprint arXiv:2307.03638 (2023). [58] Maray, Nader, et al. "Transfer learning on small datasets for improved fall detection." Sensors 23.3 (2023): 1105. [59] Chen, Siwen, et al. "Sea ice extent prediction with machine learning methods and subregional analysis in the Arctic." Atmosphere 14.6 (2023): 1023. [60] Liang, Yuxuan, et al. "A Study on Blockchain Sandwich Attack Strategies Based on Mechanism Design Game Theory." Electronics 12.21 (2023): 4417. [61] Li, Xiongquan, et al. "Unlabeled data selection for active learning in image classification." Scientific Reports 14.1 (2024): 424.