The identification and filtration of malicious texts in social media environments represent a significant technical challenge aimed at protecting users from online violence and disinformation. This complexity stems from the diversity and innovativeness of social media texts, which include unique expressions and special sentence structures. Particularly, malicious texts in interrogative forms pose alignment challenges with traditional corpora due to existing methods' failure to exploit the text's deep global semantic representations. This issue is compounded by the scant research on Chinese texts, leading to inefficiencies in recognition accuracy. To mitigate these challenges, we introduce an innovative framework based on a Global Contrastive Semantic Network (GCSN), designed to enhance malicious text recognition efficiency and accuracy by deeply learning global semantic knowledge. It comprises an encoder for global semantic information modelling and a graph-matching network for semantic similarity evaluation between question pairs, enabling the accurate identification and filtering of malicious texts with complex structures. Furthermore, we introduce a semantic consistency-based data augmentation method (COMBINE), using real-world data to generate balanced positive and negative samples, enriching the dataset and enhancing the model's ability to distinguish semantic consistency through contrastive learning. Experimental validation on two Chinese datasets demonstrates our model's exceptional performance, affirming its applicationa value in social media malicious text recognition. Our code is available at https://github.com/Wxy13131313131/GCSN-COMBINE

Current Continual Knowledge Graph Embedding (CKGE) methods primarily rely on translation-based embedding approaches, leveraging previously acquired knowledge to initialize new facts. While these methods often integrate fine-tuning or continual learning strategies to enhance efficiency, they compromise prediction accuracy and lack support for complex relational structures (e.g., multi-hop relations). To address these limitations, we propose STCKGE, a novel CKGE framework based on spatial transformation. In this framework, entity positions are jointly determined by base position vectors and offset vectors, enabling the model to represent complex relations more effectively while supporting efficient embedding updates for both new and existing knowledge through simple spatial operations, without relying on traditional continual learning techniques. Furthermore, we introduce a bidirectional collaborative update strategy and a balanced embedding method to guide parameter updates, effectively minimizing training costs while improving model accuracy. We comprehensively evaluate our model on seven public datasets and a newly constructed dataset (MULTI) focusing on multi-hop relationships. Experimental results confirm STCKGE's strong performance in multi-hop relationship learning and prediction accuracy, with an average MRR improvement of 5.4%. Our code and dataset are available at https://github.com/Wxy13131313131/STCKGE

Multi-modal emergency rumors are spreading in the current digital era, causing significant disruptions and negative impacts. Most existing methods focus on exploring rumor detection using individual small language models (SLMs) or large language models (LLMs), achieving a certain degree of success but with underlying issues. Approaches based on SLMs have reached a bottleneck due to their limited knowledge and capacity. In contrast, LLMs have unique strengths in deep analysis that compensate for the weaknesses of SLMs; however, they struggle to select and integrate analyses to draw appropriate conclusions. Furthermore, recent works on multi-modal feature fusion remain superficial, limiting the ability of these models to fully comprehend and identify rumors. In this work, we propose Collaborate Large and Small Language Models for Multi-Modal Emergency Rumor Detection (M2ERD). Specifically, it consists of two main components. First, LLMs generate multi-dimensional rationales based on multi-perspective prompts, from which SLMs selectively derive insights for rumor detection. Second, a multi-source cross-modal penetration fusion network not only accomplishes unidirectional fusion of auxiliary information such as multi-dimensional rationales but also achieves complete mutual complementation between text and the image. Comprehensive experiments demonstrate the effectiveness of M2ERD for rumor detection on Weibo, RumorEval, and Pheme datasets, achieving a 2.6% improvement in accuracy and a 1.9% improvement in F1-score compared to all baselines. We release the code and data at https://github.com/youchengyan/M2ERD.

In response to the increasing demand for efficient resource utilization in data center networks (DCNs), the development of intelligent load-balancing algorithms has become crucial. This article introduces the dual double deep Q network (D 2$$ {}<^>2 $$DQN) algorithm, designed for software-defined networking (SDN) environments within data centers. By leveraging deep reinforcement learning, D 2$$ {}<^>2 $$DQN addresses the challenges posed by dynamic traffic patterns, diverse flow requirements, and the coexistence of elephant and mice flows. Our algorithm adopts a comprehensive SDN approach, evaluating the network's status by analyzing switch load and bandwidth utilization. Using convolutional neural networks for elephant and mice flows in DCN, our algorithm enables adaptive learning and training tailored to the specific demands of elephant flows. Employing a double deep Q network architecture (DDQN), D 2$$ {}<^>2 $$DQN optimizes paths for both elephant and mice flows independently. Real-time adaptation mechanisms make routing decisions based on the robust learning capabilities of DDQN, enhancing network utilization and reducing packet loss by generating optimal forwarding paths according to the current network state and traffic patterns. Simulations conducted in a Mininet environment with RYU as the controller, utilizing a fat-tree data center topology, validate the efficacy of D 2$$ {}<^>2 $$DQN. The results demonstrate its effectiveness in achieving higher throughput, lower latency, and superior load balancing compared to traditional algorithms like equal-cost multipath and Hedera. The D 2$$ {}<^>2 $$DQN algorithm, tailored for software-defined networking in data centers, addresses challenges in dynamic traffic patterns, diverse flow requirements, and the coexistence of elephant and mice flows. Leveraging deep reinforcement learning and a dual double deep Q network architecture, it optimizes paths, achieves superior load balancing, and minimizes packet loss, outperforming traditional algorithms in throughput and latency. image

Cross-scene hyperspectral image (HSI) classification remains a challenging task due to the distribution discrepancies that arise from variations in imaging sensors, geographic regions, atmospheric conditions, and other factors between the source and target domains. Recent research indicates that convolutional neural networks (CNNs) exhibit a significant tendency to prioritize image styles, which are highly sensitive to domain variations, over the actual content of the images. However, few existing domain adaptation (DA) methods for cross-scene HSI classification take into consideration the style variations both within the samples of an HSI and between the cross-scene source and target domains. Accordingly, we propose a novel content-biased and style-assisted transfer network (CSTnet) for unsupervised DA (UDA) in cross-scene HSI classification. The CSTnet introduces a content and style reorganization (CSR) module that disentangles content features from style features via instance normalization (IN), while refining useful style information as a complementary component to enhance discriminability. A contentwise reorganization loss is designed to reduce the disparity between the separated content/style representations and the output features, thereby enhancing content-level alignment across different domains. Furthermore, we incorporate batch nuclear-norm maximization (BNM) as an effective class-balancing technique that directly exploits unlabeled target data to enhance minority class representations without requiring prior knowledge or pseudolabels, achieving better distribution alignment. Comprehensive experiments on three cross-scene HSI datasets demonstrate that the proposed CSTnet achieves state-of-the-art performance, effectively leveraging content bias and style assistance for robust DA in cross-scene HSI classification tasks. The code is available at: https://github.com/nbdszw/CSTnet.

Classifying the complicated bad information in the network can help the country grasp the distribution trend of the bad information in the network， find the work focus. At present， the research on the classification of network bad information mainly ignores the imbalance and long tailed distribution of network bad information data， making the model biased towards samples with more data during classification， resulting in a decline in overall recognition accuracy. To solve this problem， this paper proposes a classification method LTIC for long tailed bad information datasets. This method combines small sample learning with a knowledge transfer strategy. Firstly， it uses the BERT model to learn the weight of the header class， obtains the prototype of the header class through the Prototyper network specifically proposed for small sample learning， by processing the header and tail data separately to avoid data imbalance caused by training together. Then it learns the mapping relationship from the prototype to the weight， uses the learned mapping relationship to convert the prototype of the tail class into the weight of the tail class， and then connects the weight of the head class and the weight of the tail class， so as to obtain the final classification model. The experiment shows that the accuracy of this method is 80.6% and 82.3% on Twitter’s long-tail bad information dataset and THUCNews dataset， and the F1 score has significantly improved compared to the non-long tail model， effectively improving the model precision. Compared with the current relatively new long-tail dataset classification method， the average accuracy of this method is improved by 3%， with good results. Moreover， the experiment shows that when the new bad information category appears， this method can predict it with only a small amount of calculation， and the accuracy rate reaches 70%， with good scalability.

Currently, the calculation speed of the image processing algorithm is very slow when high-resolution images are used as input data.Although parallelizing the algorithm can improve its execution efficiency, the manual conversion of serial programs to parallel programs is tedious.Moreover, current automatic parallel translation tools are not scalable, and the translated program is in single parallel mode.To solve this problem, this study proposes an automatic two-level parallel translation method from C to CUDA for the Patch-based Multiple View Stereo(PMVS) algorithm, using Another Tool for Language Recognition(ANTLR) to automatically parse the source C code and identify the parallelizable loop structures by analyzing data dependencies and loop array privatization.Additionally, the loop structure of the algorithm is translated into a two-level parallel structure that includes CPU multithreading and the GPU.When the algorithm is executed, the input image is divided into two parts:one part is processed by the CPU's multithreaded code, and the other part is processed by the GPU code, thereby reducing the total execution time of the algorithm.The experimental results show that an increase in the input image resolutions gradually improves the performance of the proposed method, and the maximum speedup ratio can reach approximately 32.Moreover, the proposed method has a significantly higher speed compared with the automatic Polyhedral Parallel Code Generation(PPCG) and OpenACC translation methods.

Detection of fake news has spurred widespread interests in areas such as healthcare and Internet societies, in order to prevent propagating misleading information for commercial and political purposes. However, efforts to study a general framework for exploiting knowledge, for judging the trustworthiness of given news based on their content, have been limited. Indeed, the existing works rarely consider incorporating knowledge graphs (KGs), which could provide rich structured knowledge for better language understanding. In this work, we propose a deep triple network (DTN) that leverages knowledge graphs to facilitate fake news detection with triple-enhanced explanations. In the DTN, background knowledge graphs, such as open knowledge graphs and extracted graphs from news bases, are applied for both low-level and high-level feature extraction to classify the input news article and provide explanations for the classification. The performance of the proposed method is evaluated by demonstrating abundant convincing comparative experiments. Obtained results show that DTN outperforms conventional fake news detection methods from different aspects, including the provision of factual evidence supporting the decision of fake news detection.

Abstract Multimodal web rumors, which combine images and text, are confusing and can be inflammatory, and therefore can be harmful to national security and social stability. Currently, web rumor detection fully considers text content but ignores image content, including text embedded in images. This paper proposes a multimodal web rumor detection method based on a deep neural network considering images, image-embedded text, and text content. This method uses a VGG-19 network to extract image content features, DenseNet to extract embedded text content, and an LSTM (Long Short-term Memory) network to extract text content features. After concatenation with image features, the mean and variance vectors of the image and text shared representations are obtained through a completely connected layer, and random variables sampled from a Gaussian distribution are used to form a reparameterized multimodal feature as the input of the rumor detector. Experiments show that the accuracy of this method is 68.5% and 79.4% on Twitter and Weibo, respectively.

三维多视角立体视觉算法(patch-based multi-view stereo,PMVS)以其良好的三维重建效果广泛应用于数字城市等领域,但用于大规模计算时算法的执行效率低下.针对此,提出了一种细粒度并行优化方法,从任务划分和负载均衡、主系统存储和GPU存储、通信开销等3方面加以优化;同时,设计了基于面片的PMVS算法特征提取的GPU和多线程并行改造方法,实现了CPUs_GPUs多粒度协同并行.实验结果表明,基于CPU多线程策略能实现4倍加速比,基于统一计算设备架构(compute unified device architecture,CUDA)并行策略能实现最高34倍加速比,而提出的策略在CUDA并行策略的基础上实现了30％的性能提升,可以用于其他领域大数据处理中快速调度计算资源.

This paper analyzes the problems of excessive information resources and insufficient accuracy of data information resource representation and extraction. This paper proposes an analysis of multimedia information resource characterization extraction method based on unstructured database. First, the underlying structure of the database is optimized, and the data resource attributes are identified by combining with the data feature mining algorithm, and the low-level information resource features of the unstructured database are clustered, and the data representation extraction path is reasonably planned with reference to the information feature mining results of the underlying database. The covariance parameters in the extraction process of information resource representation are calculated with Gaussian function algorithm, so as to effectively check the extraction error of resource representation and realize the extraction method of multimedia information resource representation in unstructured database. Finally, the comparison experiment proves that the multimedia information resource representation extraction method based on unstructured database can effectively improve the efficiency and accuracy of information resource representation extraction compared with the traditional method.

In order to study the application of improved image hashing algorithm in image tampering detection, based on compressed sensing and ring segmentation, a new image hashing technique is studied. The image hash algorithm based on compressed sensing and ring segmentation is proposed. First, the algorithm preprocesses the input image. Then, the ring segment is used to extract the set of pixels in each ring region. These aggregate data are separately performed compressed sensing measurements. Finally, the hash value is constructed by calculating the inner product of the measurement vector and the random vector. The results show that the algorithm has good perceived robustness, uniqueness and security. Finally, the ROC curve is used to analyze the classification performance. The comparison of ROC curves shows that the performance of the proposed algorithm is better than FM-CS, GF-LVQ and RT-DCT.

Combining the semantic representation ability of deep learning and the discriminant ability of the significant image region, a deep representation method based on universal target detection technology is proposed. First, a general object detector is used to detect a small number of areas that are most likely to contain objects in the image. Then, the deep characterization is extracted in these areas. At the same time, in order to describe the local attributes in the region, the local invariant features are extracted and fused with the depth characterization. The fusion experiment shows that this method can achieve more abundant image characterization.

PMVS (Patch-based Multi-view Stereo) has been widely used in the 3D reconstruction,with the aerial photo of the UAV (Unmanned Aerial Vehicle).To solve the problem of the time complexity and calculation amount of PMVS,this paper proposed the two-level parallel schema of CPU multi-thread and GPU for PMVS.The solution includes GPU-based parallel design and optimization and task allocation mechanism of the images between the GPU and CPU.The experiments have been done on the platform with a 24-core CPU and NVIDIA Tesla K20 GPU high-performance server,with 16 remote sensing images havingthe resolution of 4081 X 2993.Compared with the serial traditional PMVS,the experiment results show that our model MGPS (the two-level parallel schema of CPU multi-thread and GPU for PMVS) can be 13 times faster at feature extraction,4 times faster at PMVS.Calculation error is less than 10％.MGPS shortens the execution time of PMVS algorithm.PMVS based on MGPS algorithm can also be used in the field of cultural relic protection,medical image processing,virtual reality and so on.

This paper presents an algorithm for the topic detection of food safety problems,which is using K-means clustering algorithm based on the latent dirichlet algorithm (LDA) model.The algorithm modeling the document space with LDA model,and select the probability distribution of the themes to the document as a vector of each document,process the vectors with K-means clustering algorithm,and finally get the results of the topic detection.In order to verify the effect of the test,this paper also carries out a set of traditional experiment as a control group,by using K-means algorithm based on VSM mode.Through the experiments on the data which covers 43 classes of 1920 news and Tencent micro blogging,we record the results of six experiments under different iterations and take the average.The experimental results show that the method proposed has a 20 percent increase than traditional methods on the three evaluation indices P,R and F.

Chinese abbreviations have been widely used in modern Chinese, and are one of the main sources of unknown words, resulting in the difficulty for correct Chinese character processing. This paper proposes a new approach called tr-HMM (Time Relaxed Hidden Markov Model) to recover Chinese abbreviations to their root words. The tr-HMM is a transformation of HMM, which we have devised as the basic recovery strategy. It firstly considers the abbreviation as the output sequence, and the corresponding root word as the hidden sequence. Subsequently,thetr-HMMrelaxes the time invariant hypothesis in HMM, and extends itself to a non-stationary HMM by making utilizing the time information. Lastly, we utilize the ten folds ten rounds method to prove that the tr-HMM is more effecttive than traditional recovery methods for the Chinese abbreviation recovery problem. The precision of the tr-HMM abbreviation recovery was demonstrated to be 84.3%.

One of the important works of Information Content Security is evaluating the theme words of the text. Because of the variety of the Chinese expression, especially of the abbreviation, the supervision of the theme words becomes harder. The goal of this paper is to quickly and accurately discover the intercept abbreviations from the text crawled at the short time period. The paper firstly segments the target texts, and then utilizes the Supported Vector Machine (SVM) to recognize the abbreviations from the wrongly segmented texts as the candidates. Secondly, this paper presents the collaborative methods: Improve the Conditional Random Fields (CRF) to predict the corresponding word to each character of the abbreviation; To solve the problems of the 1:n relationship, collaboratively merge the ranking list from the predict steps with the matched results of the thesaurus of abbreviations. The experiments demonstrate that our method at the recognizing stage is 76.5% of the accuracy and 77.8% of the recall rate. At the recovery step, the accuracy is 62.1%, which is 20.8% higher than the method based on Hidden Markov Model (HMM).

在深度学习理论模型的基础上,提出了基于卷积神经网络的云检测方法.以GF-2号卫星影像为数据源,选取广西壮族自治区贵港市为实验区,提取了不同下垫面的云,验证了该方法的有效性.