Home Latest Articles
Latest Articles
  • Rui Chen, Zhanquan Wang
    Journal of Computer Research and Development. 2024, 61(2): 441-459.

    To assist learners in maintaining the continuity of online learning and guiding the implementation of the optimal learning path, the intelligent tutoring system (ITS) needs to detect the tendency of learners to withdraw from learning in time and take corresponding intervention measures at the right time. Therefore, online learning session dropout prediction research is necessary. However, compared with traditional course dropouts, session dropouts occur more frequently and with shorter single study sessions. Session dropout requires the accurate prediction of learning session dropout state based on limited learning behavior feature data. Therefore, the fragmentation of learning behavior and the immediacy and accuracy of prediction results are the challenges and difficulties of learning session dropout prediction tasks. For the session dropout prediction task, we propose a unified online learning session dropout prediction model (Uni-LSDPM). Based on the pre-training and fine-tuning paradigm, Uni-LSDPM uses a multi-layer Transformer structure. In the pre-training stage, a bidirectional attention mechanism is used to learn the representation of sequence of learners’ continuous behavioral interaction features. In the fine-tuning stage, a sequence-to-sequence (Seq2Seq) attentional mechanism is used to learn the sequence combination of the learner’s continuous behavioral interaction features and dropout state. The proposed model is pre-trained and fine-tuned based on the EdNet public dataset, and the best prediction effect is obtained through ablation experiment. Comparative experiments are conducted based on multiple datasets. Experimental results show that Uni-LSDPM outperforms existing models in terms of AUC and ACC, and prove that the model has certain robustness and scalability.

  • Wanli Xue, Zhibin Zhang, Shenglei Pei, Kaihua Zhang, Shengyong Chen
    Journal of Computer Research and Development. 2024, 61(2): 460-469.

    There are three problems about feature extraction and fusion in the current mainstream tracking framework based on Transformer: 1. The two modules of feature extraction and fusion are used separately, which is easy to produce sub-optimal model training results. 2. Computational complexity of $ O\left({N}^{2}\right) $ using self-attention reduces tracking efficiency. 3. The target template selection strategy is simple and is difficult to adapt to the drastic changes in the target appearance during the tracking process. We propose a novel Transformer tracking framework using fast Fourier transform mixing target tokens and search region tokens. For problem 1, an efficient end-to-end approach is proposed to extract and fuse features for unified learning to obtain optimal model; For problem 2, the fast Fourier transform is used to achieve complete information interaction between the target tokens and search region tokens. The computational complexity of this operation is $ O\left(N\mathrm{l}\mathrm{o}\mathrm{g}\left(N\right)\right) $, which greatly improves the tracing efficiency. For problem 3, a template memory storage mechanism based on quality assessment is proposed, which can quickly adapt to the drastic changes in target appearance. Compared with the current state-of-the-art algorithms on three datasets LaSOT, OTB100 and UAV123, our tracker achieves better performance in both efficiency and accuracy.

  • Shuixiu Wu, Xianzeng Luo, Maosheng Zhong, Ruping Wu, Wei Luo
    Journal of Computer Research and Development. 2024, 61(2): 470-480.

    Learners often have problems such as “information overload” and “information trek” when facing the vast online learning course resources. Based on learners’ learning records, it is increasingly important to recommend MOOC courses to learners that are consistent with their knowledge preferences and learning needs. Aiming at the problems that the existing MOOC recommendation methods do not make full use of the implicit information contained in MOOC videos, which are easy to form a “cocoon effect”, and it is difficult to capture the dynamic learning needs and interests of learners, a dynamic MOOC recommendation model integrating video subtitle information MOOCDR-VSI is proposed, which uses BERT as the encoder to deeply mine the semantic information of MOOC video subtitle text by integrating the multi-head attention mechanism. The network based on LSTM architecture is used to dynamically capture the changing knowledge preference state of learners with learning, introduce the attention mechanism to mine the personality information and common information between MOOC videos, and finally recommend MOOC videos with Top N recall probability combined with the knowledge preference status of learners. The performance of MOOCDR-VSI is analyzed by MOOCCube in the experimental dataset collected in the real learning scenario, and the results show that the proposed model improves the HR@5, HR@10, NDCG@5, NDCG@10 , NDCG@20 evaluation indexes by 2.35%, 2.79%, 0.69%, 2.2% and 3.32%, respectively, compared with the current most optimal method.

  • Jiefan Qiu, Yifan Xu, Ruiji Xu, Dongli Zhou, Kaikai Chi
    Journal of Computer Research and Development. 2024, 61(2): 481-493.

    Because the millimeter-wave radar based on frequency-modulated continuous wave (FMCW) owns the advantages of non-contact, privacy protection, high resolution, and anti-interference, applying it in monitoring vital signs gradually becomes a trend. However, most researchers focus on how to improve the detection performance under the steady-state state and give little consideration to the detection under the non-steady state with physical disturbance. We propose a method to detect vital signs in a best-effort way. This method recognizes the steady state and non-steady state, and the type of motion obtains the vital signs without physical disturbance. For this end, we first compute feature spectrograms with range-main velocity information from motion features. And then, employ slide windows sampling to construct datasets, and train the ResNet-18 network to identify the motion state and the motion type. Finally, based on the result of the motion state, we extract the phase signal during exercise rest and use the variational mode decomposition (VMD) algorithm to analyze the respiration rate and heart rate. Experiment results show that using the trained network model, the recognition accuracy of the motion state and motion type is close to 97%, and the recognition delay is less than 1.1 s. Meanwhile, after motion recognition, the mean absolute deviation (MAD) of the detection respiration and heart rate drops to 1.7 bpm and 3.4 bpm respectively.

  • Xian Wu, Hongbo Tang, Yu Zhao, Mingyan Xu
    Journal of Computer Research and Development. 2024, 61(2): 494-502.

    The real-time migration technology of cross-cluster stateful virtual containers is an important technology in the cloud application of network services. The optimization of stateful container real-time migration technology can achieve the purpose of shortening the downtime of cloud services, which is of great practical significance in practical applications. At present, there are relatively few studies on cross-cluster container real-time migration optimization, of which iterative migration is a migration method that can effectively reduce downtime, but there are still many problems to be solved in practical applications of this technology. Firstly, a container real-time migration method is proposed for the impact of network latency on the service downtime in cross-cluster scenarios, which focuses on the network latency between clusters, and a mathematical model is established accordingly. Secondly, The influence of network latency on migration performance during cross-cluster migration is analyzed by mathematical model theory, which determines the starting conditions and the termination conditions for enabling the iterative migration process, and gives the method to calculate the number of iterations that satisfy the termination condition. The proposed model fully considers the impact of network latency on cross-cluster container real-time migration, and experimental results show that the proposed method can effectively reduce service downtime.

  • Shuping Chen, Hongmei Wei, Fei Wang, Yi Li, Wangquan He, Fengbin Qi
    Journal of Computer Research and Development. 2024, 61(2): 503-517.

    Traditional MPI (message passing interface) collectives are implemented by point-to-point messages, and have poor performance. Hardware supported collectives have attracted more and more attention due to their high performance and low CPU utilization. Aggregate tree has crucial impact on the performance of hardware supported collectives. We study the factors that affect the performance of hardware supported collectives, and propose a cost model for hardware supported collectives and an efficient method to create aggregate trees. The method includes three parts. Firstly, we choose appropriate aggregate tree type and breadth according to the operation type and the size of aggregate messages to do tradeoff between network transmission time and data processing time. Secondly, we propose a method to create hierarchical minimum height aggregate tree of type Ⅰ, which reduces the number of inter-group communication. Thirdly, we put forward a method to create the minimum cost aggregate tree of type Ⅱ, which minimizes the number of used switches. In the Sunway interconnection network, we test the proposals. In the presence of network noise, the message latency of the hierarchical minimum height aggregate tree of type Ⅰ is reduced by 24%−89% compared with that of the traditional method. The aggregate entries used by the minimum cost aggregate tree of type Ⅱ for typical communication patterns are reduced by 90% compared with that of the traditional method.

  • Linlan Liu, Zhenxing Feng, Jian Shu
    Journal of Computer Research and Development. 2024, 61(2): 518-528.

    Dynamic network link prediction has become a hot topic in network science field because of its wide application prospect. However, the complexity of spatial correlation and temporal dependence in the evolution process of dynamic network links leads to the great challenges of dynamic network link prediction task. In this paper, a dynamic network link prediction model based on sequential graph convolution (DNLP-SGC) is proposed. On the one hand, because network snapshot sequence cannot effectively reflect the continuity of dynamic network evolution, the edge trigger mechanism is employed to modify the original network weight matrix, so as to make up loss timing information in discrete snapshot of dynamic network. On the other hand, from the view of network evolution and considering the feature similarity and historical interaction information between nodes, a temporal graph convolution method is proposed to extract node features in dynamic network, and the method integrates the spatial-temporal dependence of nodes effectively. Furthermore, the causal convolutional network is used to capture the potential global temporal features in the dynamic network evolution process to achieve dynamic network link prediction. Experimental results on two real dynamic network datasets show that DNLP-SGC outperforms the baseline model on three common indexes, such as precision, recall and AUC.

  • Haijun Geng, Zhuo Meng, Shanshan Yao, Jing Yang, Haotian Chi, Xia Yin
    Journal of Computer Research and Development. 2024, 61(2): 529-538.

    Currently, intra-domain routing protocols which are deployed on the Internet respond to network failures through a global convergence mechanism. In the process of routing convergence, a large number of data packets will be discarded, leading to network interruption. The industry proposes to use routing protection methods to solve the above problem. However, the existing routing protection methods have the following four problems: 1) they cannot cope with all possible single failures in the network; 2) they need the assistance of additional auxiliary mechanisms; 3) they do not support incremental deployment; 4) each node stores multiple backup next hops for each destination, increasing storage and computation costs. We propose an intra-domain routing protection algorithm based on forwarding graph (RPBFG) to solve the above four problems of existing routing protection algorithms. We first establish a routing protection model with the goal of maximizing the failure protection ratio and the constraint that the forwarding graph contains the reverse shortest path tree; then a genetic algorithm is proposed to construct a forwarding graph that satisfies the above goal; finally, the backup next hop of all nodes arriving at the destination node is calculated according to the constructed forwarding graph. We compare the performance of RPBFG, NPC, U-turn, MARA-MA and MARA-SPE in 11 real topologies, including failure protection ratio and path stretch. The experimental results show that RPBFG can deal with all possible single failures in the network, and that is to say the failure protection ratio of RPBFG is 100%, while the average failure protection ratio of NPC, U-turn, MARA-MA and MARA-SPE is 45.05%, 74.53%, 89.78% and 91.24% respectively. In the aspect of path stretch, RPBFG decreased by 0.11%, 0.72%, 37.79% and 36.26% respectively compared with NPC, U-turn, MARA-MA and MARA-SPE. The research results will help to promote the progress of Internet service providers in deploying intra-domain routing protection schemes in real networks.

  • Zhenhua Li, Hongyi Wang, Yang Li, Hao Lin, Xinlei Yang
    Journal of Computer Research and Development. 2024, 61(1): 2-19.

    As the “last mile” of Internet content delivery, terminal networks seem rather simple but in fact constitute 99% of the performance bottlenecks. Classic design is usually oriented to typical devices and regular environments, thus making it difficult to accommodate and reproduce diversified scenarios and resulting in severe performance degradation. By comprehensively gathering and deeply diagnosing the anomalies of large-scale complex terminal networks at the cloud, we have revealed several important defects of the classic design for terminal networks in three key dimensions—availability, reliability and credibility. In order to fix these defects effectively and efficiently, the cross-layer and cross-technology collaboratively reinforced design methodology is adopted (e.g., the time-inhomogeneous 4G/5G dual connectivity management method is utilized to minimize the probability of network disconnection), so as to fulfill self-regulation mechanism design for ubiquitous scenarios. The research achievements have been applied to the high-speed network of the Ministry of Public Security, 17 million UUSpeedTest App users, 70 million Xiaomi mobile phones, 100 million Baidu PhoneGuard users, and 900 million WiFi devices. In recent years, we have also conducted forward-looking network design based on cloud-hosted emulators to discover and fix potential defects without real-world user engagement, thus making the design of terminal networks “born in the cloud and grow in the cloud”. The research achievements have been applied to Huawei DevEco Studio IDE (Integrated Development Environment), Tencent App Market, Google Android Emulator, and multiple popular Apps (like Douyin and Toutiao) of ByteDance.

  • Yu Zhang, Xinyu Jiang, Hui Yu, Jin Zhao, Hao Qi, Xiaofei Liao, Hai Jin, Biao Wang, Ting Yu
    Journal of Computer Research and Development. 2024, 61(1): 20-42.

    In recent years, some progress has been made in the key technologies of the architecture and systems software for graph processing. Large-scale graph processing has also been widely used in many fields, including scientific computing, machine learning, social networks, intelligent transportation, bioinformatics, etc. However, most real-world graph computations have characteristics such as dynamic changes and complex and diverse application requirements. This poses new demands and challenges for graph processing in terms of basic theory, architecture, and key technologies of systems software. To address these challenges, researchers have proposed a series of graph processing systems and accelerators, which optimize the graph processing process through technologies such as high-performance computing and parallel computing. Furthermore, in order to meet the demands of practical application scenarios, various graph processing frameworks and algorithms are constantly being innovated and optimized, thus enhancing the practical value of graph processing in terms of processing large-scale graph data and improving computational efficiency. We review the research and development status of key technologies in graph processing architecture and systems software, and summarize, compare, analyze the latest progress of research at home and abroad, and select fields closely related to national economy and people’s livelihood in combination with national development strategies and major application requirements. The industry progress of graph processing-related technologies is analyzed and summarized from typical applications. Finally, the future technical challenges and research directions are prospected.