Computer Science
Started in January,1974（Monthly）
Supervised and Sponsored by Chongqing Southwest Information Co., Ltd.
ISSN 1002-137X
CN 50-1075/TP
CODEN JKIEBK
Authors
Reviewers
Editors
Current Issue
Volume 48 Issue 5, 15 May 2021
  
Computer Software
Summary of Binary Code Similarity Detection Techniques
FANG Lei, WU Ze-hui, WEI Qiang
Computer Science. 2021, 48 (5): 1-8.  doi:10.11896/jsjkx.200400085
Abstract PDF(1459KB) ( 4118 )   
References | Related Articles | Metrics
Code similarity detection is commonly used in code prediction,intellectual property protection and vulnerability scan,etc.It includes source code similarity detection and binary code similarity detection.Since the source code is usually difficult to access,binary code similarity detection is more widely applicable,and a variety of detection techniques have been proposed in academia.We review researches of this field in recent years.First,we summarize the basic process of code similarity detection and challenges it faces,which include the cross-compiler,cross-optimization and cross-architecture detecting.Then,in consideration of different code information concerned,we propose to classify current binary code similarity detection techniques into 4 categories,including text-based,attribute-based measurement,program logic-based and semantic-based detection technologies,and list some representative methods and tools,such as Karta,discovRE,Genius,Gemini,SAFE,etc.Finally,according to the development context and the latest researches,we analyze and discuss the development direction of this field.
Summary on Reverse Debugging Technology
XU Jian-bo, SHU Hui, KANG Fei
Computer Science. 2021, 48 (5): 9-15.  doi:10.11896/jsjkx.200600152
Abstract PDF(1838KB) ( 1381 )   
References | Related Articles | Metrics
In the process of software development and test deployment,debugging consumes a lot of developers' energy and time.Sometimes,in order to find a critical bug,debugging needs to restart many times.Reverse debugging is a technology of software debugging.It can check running instructions and status backward without restarting,which can greatly improve the speed of software debugging,reduce the difficulty of software development,and effectively repair the errors during the running of the program.The core issue of this technology is the recovery of running state.At present,current solutions are state preservation and state reconstruction.This paper mainly reviews the development of reverse debugging from the aspects of principle,academic research,products implementation and technical application.It focuses on the time-based and instruction-based state preservation and two methods of reverse execution reconstruction states.It summarizes 3 specific applications of record replay program execution,location analysis software error,reverse data flow recovery,which provides a reference for the research and application of reverse debugging technology.
Test Case Generation Method Oriented to Tabular Form Formal Requirement Model
WANG Wen-xuan, HU Jun, HU Jian-cheng, KANG Jie-xiang, WANG Hui, GAO Zhong-jie
Computer Science. 2021, 48 (5): 16-24.  doi:10.11896/jsjkx.201000048
Abstract PDF(1897KB) ( 1159 )   
References | Related Articles | Metrics
The rapid growth of the software size and complexity of modern safety-critical systems has brought many challenges to the development of such safety-critical software systems.Traditional text documents cannot guarantee the development progress and system reliability requirements.For this reason,this paper proposes a formal form requirement modeling method with both readability and automatic analysis.This paper introduces a method for automatically generating test cases for this tabular model.The work includes semantic analysis of the formal requirements tabular model,establishing the control tree structure of the requirements model,and obtaining its test equivalence classes.In order to reduce unnecessary testing,test path constraint selection methods are proposed based on those criteria.Through performing domain error test case selection,test cases are generated for each path constraint selected,which makes up a test case set for the requirement.At last,to demonstrate how we generate test cases form a requirement model,a case study is given.
SymFuzz:Vulnerability Detection Technology Under Complex Path Conditions
LI Ming-lei, HUANG Hui, LU Yu-liang, ZHU Kai-long
Computer Science. 2021, 48 (5): 25-31.  doi:10.11896/jsjkx.200600128
Abstract PDF(2058KB) ( 1375 )   
References | Related Articles | Metrics
The current vulnerability detection technology can realize the rapid detection of small-scale programs,but it is inefficient when performing vulnerability detection on programs with large or complex path conditions.In order to achieve a rapid detection of vulnerabilities under complex path conditions,this paper proposes a vulnerability detection technology SymFuzz under complex path conditions.SymFuzz combines guided fuzzing technology and selected symbolic execution technology,filters program paths through guided fuzzing technology,and uses selected symbolic execution technology to solve paths that may trigger vulnerabilities.This technology first obtains program vulnerability information through static analysis.Then it uses guided fuzzy test technology to quickly generate test cases that can cover the vulnerability function.Finally,it executes symbolic execution on the path that can trigger the vulnerability within the vulnerability function to generate a test case that triggers the program vulnerability.This paper implements the prototype system of SymFuzz based on open source projects such as AFL and S2E.The comparison experiments show that SymFuzz significantly improves the effectiveness of vulnerability detection under complex path conditions compared with existing fuzzy testing techniques.
Terminology Recommendation and Requirement Classification Method for Safety-critical Software
YANG Zhi-bin, YANG Yong-qiang, YUAN Sheng-hao, ZHOU Yong, XUE Lei, CHENG Gao-hui
Computer Science. 2021, 48 (5): 32-44.  doi:10.11896/jsjkx.210100105
Abstract PDF(3149KB) ( 1656 )   
References | Related Articles | Metrics
Most of the knowledge in the requirements of safety-critical software needs to be manually extracted,which is time-consuming and laborious.Recently,artificial intelligence technology has been gradually used in the design and development of safety-critical software,to reduce the work of engineers and shorten the life cycle of software development.This paper proposes a terminology recommendation and requirement classification method for safety-critical software.Firstly,the terminology recommendation method extracts candidate terms based on part-of-speech rules and dependency rules and clusters candidate terms through term similarity calculation.The clustering results are recommended to engineers.Secondly,the requirement classification method automatically classifies safety-critical software requirements as functional,safety,reliability,etc.based on feature extraction.Finally,the prototype tool TRRC4SCSTool is implemented in the AADL open-source modeling environment OSATE,and the experimental analysis is carried out through the dataset collected from the industrial requirements and safety certification standards,and the results show the effectiveness of the method.
Data-driven Methods for Quantitative Assessment and Enhancement of Open Source Contributions
FAN Jia-kuan, WANG Hao-yue, ZHAO Sheng-yu, ZHOU Tian-yi, WANG Wei
Computer Science. 2021, 48 (5): 45-50.  doi:10.11896/jsjkx.201000107
Abstract PDF(1723KB) ( 1596 )   
References | Related Articles | Metrics
In recent years,open source technologies,open source software and open source communities have become increasingly significant in digital era,and it has become an important trend to study the open source field through quantitative analysis me-thods.Developers are the core of open source projects,and the quantification of their contributions and the strategies to improve their contributions after quantification are the key to the healthy development of open source projects.We propose a data-driven method for quantitative assessment and continuous optimization of open source contributions.Then,we implement it through a practical framework,Rosstor (Robotic Open Source Software Mentor).The framework consists of two main parts.One is a contribution evaluation model,it adopts an entropy-weight approach and can dynamically and objectively evaluate developers' contributions.Another is a model to enhance contributions,it adopts a deep reinforcement learning approach and can maximize develo-pers' contributions.Contributors' data from a number of famous open source projects on GitHub are selected,and through massive and sufficient experiments,it verifies that Rosstor not only makes the developers' contributions on all projects to be greatly improved,but also has a certain degree of immunity,which fully proves the effectiveness of the framework.The Rosstor framework provides methodological and instrumental support for the sustainable health of open source projects and the open source community.
Class Flattening Method for AltaRica 3.0 Model
QI Jian, HU Jun, GU Qing-fan, RONG Hao, ZHAN Wan-li, DONG Yan-hong
Computer Science. 2021, 48 (5): 51-59.  doi:10.11896/jsjkx.200700184
Abstract PDF(2982KB) ( 688 )   
References | Related Articles | Metrics
AltaRica is a modeling language for complex safety-critical systems.Guarded Transition System(GTS) is the latest exe-cution semantic model of AltaRica 3.0.The flattening of classes in the AltaRica 3.0 hierarchical syntax model is an important step in the process of transforming the AltaRica 3.0 syntax model into an equivalent flattened GTS semantic model.In this paper,a flattening optimization method for classes in AltaRica 3.0 models is proposed.Firstly,this paper designs a dedicated data structure to store the semantic structure of the class in the AltaRica 3.0 models,refines and defines the granularity of the AltaRica 3.0 model described by the original ANTLR(Another Tool for Language Recognition) meta language.Secondly,this paper generates the corresponding lexical and syntax analyzer based on ANTLR to automatically construct the syntax tree of the input model.Through traversing the syntax tree,the key information of fine-grained class is obtained and stored.Then,a dedicated algorithm is designed to realize the flattening process of the class efficiently.Finally,the correctness and effectiveness of this me-thod are verified through the analysis of several example systems.
Black-box Adversarial Attack Method Towards Malware Detection
CHEN Jin-yin, ZOU Jian-fei, YUAN Jun-kun, YE Lin-hui
Computer Science. 2021, 48 (5): 60-67.  doi:10.11896/jsjkx.200300127
Abstract PDF(2755KB) ( 1358 )   
References | Related Articles | Metrics
Deep learning method has been widely used in malware detection,which also has an excellent performance in the aspect of classification accuracy.Meanwhile,deep neural networks are vulnerable to adversarial attacks in the form of subtle perturbations added on the input data,resulting in incorrect predictive results,such as escaping the malware detection.Aiming at the security of malware detection method based on deep learning,this paper proposes a black-box adversarial attack method towards the malware detection model.First,it uses the generative adversarial net model to generate the adversarial examples.Then,the gene-rated adversarial examples are identified as the pre-set target type to achieve the target attack.Finally,experiments are carried out on the Kaggle competition malware dataset to verify the effectiveness of the black-box attack method.Furthermore,the generated adversarial examples are applied to attack other classification models to testify the strong transfer attack capacity of the proposed black-box attack method.
Tree Structure Evaluation Visualization Model for Program Debugging
SU Qing, LI Zhi-zhou, LIU Tian-tian, WU Wei-min, HUANG Jian-feng, LI Xiao-mei
Computer Science. 2021, 48 (5): 68-74.  doi:10.11896/jsjkx.200100133
Abstract PDF(3900KB) ( 963 )   
References | Related Articles | Metrics
Tree structure is a nonlinear data structure,whose evolution process is abstract in the program execution.Memory leak happens while applying modification-like operation on it.It is an open challenge for programming beginner to control the evolution of tree structure in the program debugging procedure,especially to debug the error when memory leakage happens.To address this issue,this paper proposes a tree evaluation visualization model(TEVM) to visualize the evolution procedure of tree structure.Two tree structures are obtained after executing a visual debugging step.This paper designs a structure comparison algorithm to obtain their structural difference including leaked tree by transforming tree structure into linear representation.It also designs a tree layout method and computes their positional difference.A visual evaluation sequence is generated with thesestructural diffe-rence and positional difference.At last,it applys the drawing engine to interpret and execute actions of this sequence to visualize the tree structure evaluation dynamically,smoothly and intuitively.The visualization of tree structure helps programming beginner to understand the execution of program relating to tree structure and improves the efficiency of program debugging.TEVM model is applied in the Web AnyviewC,which is a prototype of the integrated development environment for programming trai-ning,and gains an excellent application effect.
Empirical Study on Stability of Clone Code Sets Based on Class Granularity
ZHANG Jiu-jie, CHEN Chao, NIE Hong-xuan, XIA Yu-qin, ZHANG Li-ping, MA Zhan-fei
Computer Science. 2021, 48 (5): 75-85.  doi:10.11896/jsjkx.200900062
Abstract PDF(2539KB) ( 767 )   
References | Related Articles | Metrics
Researching on clone code is closely related to various problems in software engineering.The existing researches and studies on stability of clone code mainly focus on comparisons between clone code and non-clone code,or between different types of clone code.Rare studies consider the object-oriented classes in which clone sets distribute.This paper presents a comprehensive empirical study on stability of clone sets based on object-oriented class granularity.This paper frames four study problems about the stability of clone sets.Around these particular problems,all clone sets are categorized into three groups,intra-class clone sets,inter-class clone sets and hybrid-class clone sets.And stability of them is compared and analyzed by 9 evolution patterns from 4 perspectives during the process of software evolution.First of all,clone code fragments in all revisions of subject systems are detected and tagged with object-oriented classes where they distribute in.Next,clone sets between adjacent revisions are mapped based on mapping clone fragments,and evolution patterns of clone sets can be recognized and tagged.After that,clone genealogy is constructed by combing the results of mapping relations and evolution patterns,and then stability of three groups of clone sets is calculated from different perspectives.Eventually,differences of three groups are compared and analyzed.According to the experimental results on 7 700 revisions of seven diverse object-oriented subject systems,about 60% of intra-class clone sets have a life cycle more than half of the total number of reversions,the percentage of inter-class clone sets and hybrid clone sets that have a life cycle rate of 50% or more are both close to 35%.Comparatively speaking,among three kinds of clone sets,the frequency of changes within intra-class clone sets is the lowest.Also,there is a bit more merging,branching and late propagation evolution patterns in inter-class clone sets.And the frequency of fragments deletions,consistent changes and inconsistent changes is the highest in hybrid-class clone sets.Overall,stability of intra-class clone sets is the best,hybrid-class clone sets should be given a higherpriority to tracing or refactoring in the process of software evolution.The clone code stability analysis methods and findings from this work will provide a strong reference and support for clone code maintenance,tracking,refactoring and other cloning management related software activities.
Software Reliability Prediction Based on Continuous Deep Confidence Neural Network
QI Hui, SHI Ying, LI Deng-ao, MU Xiao-fang, HOU Ming-xing
Computer Science. 2021, 48 (5): 86-90.  doi:10.11896/jsjkx.210200055
Abstract PDF(1761KB) ( 900 )   
References | Related Articles | Metrics
In order to improve the accuracy of intelligent prediction of software reliability,continuous depth confidence neural network algorithm is used for software reliability prediction.Firstly,the core elements samples that affect software reliability are extracted,and the key features of the sample elements are obtained.Then,a software reliability prediction model based on conti-nuous deep belief neural network (DBN) is established.The samples to be predicted are input,and the parameters such as DBN weight are obtained through pre-processing training of multiple Restricted Boltzmann Machine (RBM) layers and multiple reverse fine-tuning iterations until the maximum number of RBM layers and the maximum number of reverse fine-tuning iterations are reached.Finally,a stable software reliability prediction model is obtained.Experiments show that good software reliability prediction accuracy and standard deviation can be obtained by reasonably setting the number of nodes in the hidden layer of DBN and the learning rate.Compared with commonly used software reliability prediction algorithms,this algorithm has high prediction accuracy,small standard deviation and high applicability in software reliability prediction.
Approach for Knowledge-driven Similar Bug Report Recommendation
YU Sheng, LI Bin, SUN Xiao-bing, BO Li-li, ZHOU Cheng
Computer Science. 2021, 48 (5): 91-98.  doi:10.11896/jsjkx.200600159
Abstract PDF(1736KB) ( 796 )   
References | Related Articles | Metrics
Software bug is inevitable in the process of software development,and the submitted bug reports are important source of information for bug analysis and fixing.Developers usually refer to similar historical bug reports and fixing solutions to analyze and fix the new bug at hand.This paper proposes an approach for knowledge-driven similar bug report recommendation.Based on the combination of information retrieval and Word Embedding,it constructs a bug knowledge graph.Then,it uses TF-IDF and Word Embedding to calculate the text similarity between bug reports,and generates the similarity of primary and secondary attributes between the bug reports.Finally,the above similarities are merged into a comprehensive similarity,and similar bug reports are recommended based on the comprehensive similarity.Experimental results show that the proposed approach improves the performance by an average of 12.7% on the Firefox dataset compared to the baseline method.
Test Case Prioritization Combining Clustering Approach and Fault Prediction
XIAO Lei, CHEN Rong-shang, MIAO Huai-kou, HONG Yu
Computer Science. 2021, 48 (5): 99-108.  doi:10.11896/jsjkx.200400100
Abstract PDF(4745KB) ( 961 )   
References | Related Articles | Metrics
The rapid delivery of software leads to the frequent execution of regression testing.The higher efficiency of regression testing is required for the quick fault-feedback in Continuous integration(CI).The goal of test case prioritization(TCP) approach is that the test cases with the higher fault detection rate are preferentially executed.Therefore,TCP approach meets the requirement of quick fault-feedback in CI.The fault prediction approach can predict the failed probability in the new version using the code feature and the historical failure information.The choice of the number of cluster and the feature subset are not considered in the traditional clustering TCP approaches.This paper proposes a test case prioritization method combining clustering approach and fault prediction,which fisrtly identifies the correlation between the test cases and the codes,then divides the test cases into the different clusters,lastly implements the inter-cluster and intra-cluster prioritization on the guidance of the fault prediction and the maximum and minimum distance strategy.The experimental results verify that the efficiency of prioritization is influenced by the choice of the number of cluster and the feature subset.If the best clustering number and the feature subset are not required,the proposed approach is superior to the traditional clustering TCP approaches.
Database & Big Data & Data Science
Combining User-end and Item-end Knowledge Graph Learning for Personalized Recommendation
LIANG Hao-hong, GU Tian-long, BIN Chen-zhong, CHANG Liang
Computer Science. 2021, 48 (5): 109-116.  doi:10.11896/jsjkx.200600115
Abstract PDF(1912KB) ( 1077 )   
References | Related Articles | Metrics
How to accurately model user preferences based on existing user behavior and auxiliary information is of great important.Among all kinds of auxiliary information,Knowledge Graph (KG) as a new type of auxiliary information,its nodes and edges contain rich structural information and semantic information,and attracts a growing researchers' attention in recent years.Plenty of studies show that the introduction of KG in personalized recommendation can effectively improve the performance of recommendation,and enhance the rationality and interpretability of recommendation.However,the existing methods either explore the independent meta-paths for user-item pairs over KG,or adopt graph representation learning on whole KG to obtain representations for users and items separately.Although both have achieved certain effects,the former fails to fully capture the structural information of user-item pairs in KG,while the latter ignores the mutual effect between target user and item during the embedding propagation.In order to make up for the shortcomings of the above methods,this paper proposes a new model named User-end and Item-end Knowledge Graph (UIKG),which can effectively capture the correlation between users' personalizedpreferences and items by mining the associated attribute information in their respective KG.Specifically,we learn the user representation vectors from the user KG,and then introduce the user representation vectors into the item KG based on the method of graph convolution neural network to jointly learn the item representation vectors,so as to realize the seamless unity of the user KG and the item KG.Finally,we predict the user preference probability of the item through MLP.Experimental results on open datasets show that,compared with the baseline method,UIKG improves by 2.5%~13.6% on Recall@K index,and 0.4%~5.8% on AUC and F1 indexes.
Social Rumor Detection Method Based on Multimodal Fusion
ZHANG Shao-qin, DU Sheng-dong, ZHANG Xiao-bo, LI Tian-rui
Computer Science. 2021, 48 (5): 117-123.  doi:10.11896/jsjkx.200400057
Abstract PDF(1942KB) ( 1591 )   
References | Related Articles | Metrics
With the development of social networking platforms,social networks have become an important source of information for people.However,the convenience of social networks has also led to the rapid propagation of false rumors.Compared with textual rumors,social network rumors with multimedia content are more likely to mislead users and get dissemination,so the detection of multi-modal rumors is of great significance in real life.Several multi-modal rumor detection methods have been proposed,but the visual features and joint representation of text and visual features have not been fully explored in current approaches.To make up for these shortcomings,an end-to-end multi-modal fusion network based on deep learning is developed.Firstly,the visual features of each region of interest in the image are extracted.Then,the text and the visual features are updated and fused by using a multi-head attention mechanism.Finally,these features are concatenated based on the attention mechanism for the detection of multi-modal rumors in social networks.Comparative experiments on the public data sets of Twitter and Weibo are conducted and experimental results show that the proposed method has a 13.4% F1 value increase on Twitter data set and a 1.6% F1 value increase on Weibo data set.
Post-processing Network Embedding Algorithm with Random Projection and Principal Component Analysis
HU Xin-tong, SHA Chao-feng, LIU Yan-jun
Computer Science. 2021, 48 (5): 124-129.  doi:10.11896/jsjkx.200500058
Abstract PDF(1627KB) ( 802 )   
References | Related Articles | Metrics
Network embedding as network representation learning has received a lot of attention from researchers in recent years.A number of models based on low-dimensional vector representation of nodes in network structure learning networks,such as DeepWalk,have been developed with good results in tasks such as node classification and link prediction.However,with the network size increases,there are computational bottlenecks with multiple network embedding algorithms.To mitigate this problem,no-learning methods such as random projection can be used,but critical information about the network structure may be lost,resulting in degraded algorithm performance.In this paper,a post-processing algorithm for network embedding(PPNE) is proposed,which uses random projection as well as principal component analysis to effectively retain key information and maintain a higher order approximation of the network structure.Experiments are conducted on three public datasets for node classification and link prediction tasks,while the performance of the PPNE algorithm is verified against other network embedding algorithms.The experimental results show that the PPNE algorithm has a large improvement over other algorithms in terms of both perfor-mance and running time,and the algorithm has a speed improvement of at least two orders of magnitude over other learning-based algorithms while ensuring good task performance.
Study on Predictive Erasure Codes in Distributed Storage System
ZHANG Hang, TANG Dan, CAI Hong-liang
Computer Science. 2021, 48 (5): 130-139.  doi:10.11896/jsjkx.200300124
Abstract PDF(2390KB) ( 856 )   
References | Related Articles | Metrics
Erasure coding consumes less storage space and obtains a higher data reliability,thus being widely used by distributed storage systems.However,when erasure codes are used to repair data,their high repair costs limit their application.In order to reduce the repair cost of erasure codes,researchers have researched a lot on block codes and regenerative codes.But block codes and regeneration codes are passive fault tolerance.For some nodes that are prone to failure,using active fault tolerance can better reduce repair costs and maintain the system reliability.Therefore,this paper proposes a proactive basic-Pyramid(PPyramid) code.The PPyramid code uses the hard disk failure prediction method to adjust the association between redundant and data blocks in the Pyramid code,divides hard disks that are predicted to fail into the same group,thus making all read operations to be performed within the team when recovering data,thereby reducing the number of read data blocks and saving repair costs.In a distributed storage system based on Ceph,it is compared with other commonly used erasure codes,when repairing multiple hard drives.Experimental results show that,PPyramid codes can reduce repair costs by 6.3%~34.9% and decrease repair time by 7.6%~63.6% compared with basic-Pyramid.Compared with LRC code,pLRC code,SHEC code and DLRC code,it can reduce repair costs by 8.6%~52% and decrease repair time by 10.8%~52.4%.Meanwhile,PPyramid codes are flexible in construction and have strong practical application value.
Importance Evaluation Algorithm Based on Node Intimate Degree
MA Yuan-yuan, HAN Hua, QU Qian-qian
Computer Science. 2021, 48 (5): 140-146.  doi:10.11896/jsjkx.200300184
Abstract PDF(4749KB) ( 1173 )   
References | Related Articles | Metrics
Identifying important nodes in complex networks has been a hot topic in the field of social network analysis and mi-ning,which helps to understand the role of influential communicators in information diffusion and the spread of infectious diseases.The existing algorithm of node importance takes neighbor information into account,but ignoring the structure information between node and neighbor node.To solve this problem,considering the different influence of the neighbor node to node under different structures,this paper proposes a node-importance evaluation algorithm that takes into account the number of neighbors of a node and the intimacy between nodes and neighbors,which embodies the degree of node and “intimate” attribute.In this algorithm,similarity index is used to measure the intimacy between nodes,and Kendall correlation coefficient is used to evaluate the accuracy of node ranking.The SIR(susceptible-infected-recovered) model is used to simulate the propagation process on several classical networks.The results show that compared with degree index,closeness centrality index,betweenness centrality index and K-shell index,KI index can rank the propagation influence of nodes more accurately.
Collaborative Filtering Recommendation Algorithm Based on User Rating and Similarity of Explicit and Implicit Interest
WU Jian-xin, ZHANG Zhi-hong
Computer Science. 2021, 48 (5): 147-154.  doi:10.11896/jsjkx.200300072
Abstract PDF(1850KB) ( 768 )   
References | Related Articles | Metrics
Collaborative filtering algorithm is the most widely used algorithm in recommendation system.Its core is to use a group with similar interests to recommend information of interest for users.The traditional collaborative filtering algorithm uses the user item scoring matrix to calculate the similarity,and finds the similar groups of users through the similarity to recommend.However,due to the sparsity of the scoring matrix,the calculation of the similarity is not accurate enough,which indirectly leads to the degradation of the quality of the recommendation system.In order to alleviate the impact of data sparsity on the similarity calculation and improve the quality of recommendation,a similarity calculation method is proposed,which integrates user rating and user's explicit and implicit interest.This method first uses the user item scoring matrix to calculate the similarity of user's scoring,then uses the basic attribute of user and the user item scoring matrix to get the implicit attribute of project,then integrates the attribute of project category,the implicit attribute of project,the scoring matrix of user item and the scoring time of user to get the similarity of user's explicit and implicit interest,finally integrates the similarity of user's scoring and the similarity of user's explicit and implicit interest to find similar groups of users for recommendation.Experimental results on the data set Movielens show that compared with the traditional algorithm,which only uses a single scoring matrix to calculate the similarity,the new similarity calculation method can not only find similar groups of users more accurately,but also provide a better recommendation quality.
Computer Graphics & Multimedia
HEVC Post-processing Algorithm Based on Non-local Low-rank and Adaptive Quantization Constraint Prior
XU Yi-fei, XIONG Shu-hua, SUN Wei-heng, HE Xiao-hai, CHEN Hong-gang
Computer Science. 2021, 48 (5): 155-162.  doi:10.11896/jsjkx.200800079
Abstract PDF(1619KB) ( 653 )   
References | Related Articles | Metrics
Video compressed by HEVC has an obvious compression effect under the condition of a high compression ratio and a low bit rate.To solve this problem,a post-processing algorithm of HEVC based on non-local low-rank (NLLR) and adaptive quantization constraint (AQC) prior is proposed.This algorithm firstly constructs the optimization problem within the maximum priori probability framework.Then,the decoded compressed video and quantization parameters QP are used to obtain the NLLR and AQC prior information.Finally,the split-Bregman iterative algorithm is used to solve the optimization problem,so as to effectively remove the compression effect and improve the quality of reconstructed video.Among them,the NLLR prior is obtained by constructing the non-local low-rank model based on similar-block clustering.The AQC prior is obtained by combining the constraint characteristics under different quantization parameters QP and the DCT domain block activity of video.Experimental results show that the proposed algorithm can achieve an average PSNR improvement of 0.259 7 dB in intra-frame coding mode and an average PSNR improvement of 0.282 8 dB in inter-frame coding mode compared with HEVC standard at the same bit rate.
Detection of Head-bowing Abnormal Pedestrians Based on Human Joint Points
GUAN Wen-hua, LIN Chun-yu, YANG Shang-rong, LIU Mei-qin, ZHAO Yao
Computer Science. 2021, 48 (5): 163-169.  doi:10.11896/jsjkx.200800214
Abstract PDF(5136KB) ( 1303 )   
References | Related Articles | Metrics
In recent years,with the rapid development of smart phones,head-bowing pedestrians keep browsing mobile phones when they cross the road.As a result,they often cause the traffic accidents.How to effectively detect the bow-headed people has become an urgent problem.The existing detection methods not only require a large number of datasets about real pedestrians using mobile phones,but also have the problems of low recognition accuracy and unsatisfactory speed.Considering these problems,this paper proposes a fast and effective method to detect head-bowing abnormal pedestrians.This method is based on joint points instead of images,therefore it is different from existing methods.Firstly,we propose a novel method to construct a synthetic dataset,which solves the problem of lacking dataset and corresponding labels.We leverage human joint points by adjusting the coordinates of the left and right hand to simulate the posture of device-holding.Secondly,we achieve the abnormal behavior recognition and classification by training network on our synthetic dataset.Additionally,we make full use of arm and head information to achieve a precise pedestrian abnormal behavior detection.Finally,experiment results show that the proposed method can achieve real-time detection,and the detection accuracy reaches 94.08%.Therefore,it can provide necessary reference information for video surveillance,drivers,assisted driving,and autonomous driving systems.
Motion-estimation Based Space-temporal Feature Aggregation Network for Multi-frames Rain Removal
MENG Xiang-yu, XUE Xin-wei, LI Wen-lin, WANG Yi
Computer Science. 2021, 48 (5): 170-176.  doi:10.11896/jsjkx.210100104
Abstract PDF(4137KB) ( 1079 )   
References | Related Articles | Metrics
Outdoor videos obtained under rainy weather cause visual quality degradation,which affects the processing effects of visual tasks such as object recognition and tracking.In order to enhance the quality of video and complete the effective recovery of the details in the motion video,many methods have been proposed in video rain removal.At this stage,most of the video rain removal methods based on convolutional neural networks employ single-frame enhancement and multi-frame fusion to remove rain.But the movement of some pixels between adjacent frames with direct enhancement is difficult to be completed in the temporal dimension.And the manner cannot effectively achieve end-to-end training,making the final result still relatively blurry and many detailed information losses.In order to effectively solve the above problems,this paper proposes a multi-frame fusion rain removal network based on the combination of motion estimation and space-temporal feature aggregation,ME-Derain for short.First,the optical flow estimation method is used to establish a reference frame to complete the alignment of adjacent frames,and then an encoder-decoder structure is introduced.The convolutional neural network connected by the residual connection and the time-related attention enhancement mechanism together form a multi-frame fusion network.Finally,the enhancement module related to the spatial sequence is used to obtain the rain removal video.A large number of experiments on different data sets show that the proposed method is better than most common methods at this stage and can obtain better rain removal effect.
Background-aware Correlation Filter Tracking Algorithm with Adaptive Scaling and Learning Rate Adjustment
CHEN Yuan, HUI Yan, HU Xiu-hua
Computer Science. 2021, 48 (5): 177-183.  doi:10.11896/jsjkx.200300109
Abstract PDF(4018KB) ( 704 )   
References | Related Articles | Metrics
Aiming at the problem of object tracking drift caused by occlusion factors and target scale changes during the tracking process,this paper proposes an adaptive scale and learning rate-adjusted background-aware correlation filter tracking algorithm.First,this algorithm obtains the target's initial position information through the background-aware correlation filter;then,it trains the scale correlation filter under the basic framework of the background-aware correlation filter and estimates the target scale change effectively,thus accurately adjusting the search area size;next,the occlusion determination is performed according to the fluctuation of the response map,and the average peak energy index and the maximum response value are used to estimate target occlusion,thus enabling the model to adaptively adjust learning rate;finally,this algorithm designs the corresponding model update strategy to improve the model performance.This algorithm is tested on the OTB100 Benchmark dataset,and test result show that this algorithm improves the success rate by 6.2% and the accuracy by 10.1% compared with the background-aware correlation filter.Therefore,the proposed algorithm can effectively deal with occlusion and scale changes,improve the success rate and accuracy of the tracking model,and have a real-time tracking speed.
Artificial Intelligence
Autonomous Structural Adjustment of Crowd Intelligence Network: Begin from Structure of Biological Regulatory Network
YIN Zi-qiao, GUO Bing-hui, MA Shuang-ge, MI Zhi-long, SUN Yi-fan, ZHENG Zhi-ming
Computer Science. 2021, 48 (5): 184-189.  doi:10.11896/jsjkx.210200161
Abstract PDF(3215KB) ( 685 )   
References | Related Articles | Metrics
As one of the most important research directions in the artificial intelligence 2.0 era,crowd intelligence has received extensive attention from researchers in the industry and academia.Traditional artificial intelligence models tend to use the fully connected network structure to achieve higher accuracy.However,in a complex confrontation environment with stronginterfe-rence,the intelligent decision-making system needs to face system structural perturbations caused by communication interference or even targeted attack.Without losing too much accuracy,in order to achieve the demand for faster and more stable real-time response,it is necessary for intelligent system to have a real-time autonomous response structural adjustment mechanism.Such autonomous corresponding adjustment mechanisms are common in regulatory networks for biological systems.By introducing DReSS index family,this research quantitatively analyzes the impact of structural perturbations on state spaces in random and real networks.The anti-interference feature of different network structures against structural perturbations is compared.An autonomous adjustment concepts for the network structure of the crowd intelligence systems is proposed in this research.
Humans-Cyber-Physical Ontology Fusion of Industry Based on Representation Learning
YANG Ru-han, DAI Yi-ru, WANG Jian, DONG Jin
Computer Science. 2021, 48 (5): 190-196.  doi:10.11896/jsjkx.200500023
Abstract PDF(2764KB) ( 795 )   
References | Related Articles | Metrics
With the development of the new generation of artificial intelligence technology,the manufacturing system has deve-loped from the humans-physical binary system to the humans-cyber-physical triad system,and the multivariate data fusion across domains and layers has become an inevitable trend.As a conceptual model capable of describing data semantically,ontologies are widely used for integration,sharing and reuse of multiple heterogeneous data.In traditional industries,research on using ontology fusion to drive data fusion typically focuses on information and physical systems.In order to solve the problem of humans-cyber-physical ontology fusion,an improved representation learning model TransHP is proposed in this paper.The classical translation model does not effectively use information other than the structure of the triad,TransHP makes improvements based on TransH,and elements in the ontology constitute the type triad and the instance triad.In the TransHP,first,for the type triad,the structure and properties of the triad are used for joint training.Next,the vector representations of the obtained type entities are used as the input of the training instance vectors,and the structure information of the instance is used for joint training,while the confidence level is added as the relational strength calculation to solve the problem of the disorderly distribution of entities in semantic space caused by the sparsity of the relational triad.In this paper,a human subject ontology is constructed as an example of a hot-rolling production process in an industrial field and tested as a small sample,and the experimental result shows that the TransHP model infers a richer and more accurate relationship between entities in the human subject ontology compared with the TransH model.The fusion of humans-cyber-physical ontologies has been realized through the TransHP model,which solves the problem of humans-cyber-physical information interaction and paves the way for collaborative decision-making.
Wide and Deep Learning for Default Risk Prediction
NING Ting, MIAO De-zhuang, DONG Qi-wen, LU Xue-song
Computer Science. 2021, 48 (5): 197-201.  doi:10.11896/jsjkx.200900043
Abstract PDF(1591KB) ( 1387 )   
References | Related Articles | Metrics
Default risk control is a key business component of credit loan services,which directly affects the profitability and bad-debt rate of lenders.With the development of the mobile Internet,credit-based financial services have benefited the general public.Default risk control has changed from manual judgment based on rules to credit models built by using large amounts of customer data to predict the default rate of customers.Relevant models include traditional machine learning models and deep learning mo-dels.The former has a strong interpretability but a weak predictability;the latter has a strong predictability but a poor interpre-tability,which is prone to overfitting the training data.Therefore,the integration of traditional machine learning models and deep learning models has always been an active research area in credit modeling.Inspired by the wide & deep learning models in re-commendation systems,a credit model first can utilize traditional machine learning to capture features of the structured data,while a deep learning can capture features of the unstructured data.Then,the model combines two parts of the learned features and uses an additional linear layer to transform the hidden features.Finally,the model outputs the predicted default rate.This model neutralizes the advantages of traditional machine learning models and deep learning models.Experimental results show thatthe proposed model has a stronger capability to predict the default probability of customers.
Chinese Event Detection with Hierarchical and Multi-granularity Semantic Fusion
DING Ling, XIANG Yang
Computer Science. 2021, 48 (5): 202-208.  doi:10.11896/jsjkx.200800038
Abstract PDF(2068KB) ( 1004 )   
References | Related Articles | Metrics
Event detection is an important task in information extraction field,which aims to identify trigger words in raw text and then classify them into correct event types.Neural network based methods usually regard event detection as a word-wise classification task,which suffers from the mismatch problem between words and triggers when applied to Chinese.Besides,due to the multiple word senses of a trigger word,the same trigger word in different sentences causes the ambiguity problem.To address the two problems in Chinese event detection,we propose a Chinese event detection model with hierarchical and multi-granularity semantic fusion.First,we adopt a character-based sequence labelling method to solve the mismatch problem,in which we devise a Character-Word Fusion Gate to capture the semantic information of words in different segmentation ways.Then we device a Character-Sentence Fusion Gate to learn a character-word-sentence hybrid representation of sequence,which takes the semantic information of the entire sentence into condition and solves the ambiguity problem.Finally,in order to balance the influence the label “O” and the other labels,a loss function with bias is applied to train our model.The experimental results on the widely used ACE2005 dataset show that our approach outperforms at least 3.9%,1.4% and 2.9% than other Chinese event detection models under the metrics of accuracy (Precision,P),recall (Recall,R) and F1.
Trial Risk Assessment System Based on Fuzzy Number Similarity
YONG Qi, JIANG Wei-na, LUO Yu-ze
Computer Science. 2021, 48 (5): 209-216.  doi:10.11896/jsjkx.200500034
Abstract PDF(2172KB) ( 713 )   
References | Related Articles | Metrics
As the trial management becomes more refined,the demand of courts for accurate assessment of trial risk is gradually increasing.How to effectively carry out quantitative analysis of trial risk level is the core issue of precise risk assessment.Existing methods such as machine learning and fuzzy matrix are difficult to be effectively applied to actual trial scenes due to limited historical data and high accuracy of trial risk assessment.In response to this,this paper builds the trial risk assessment model based on the fuzzy number similarity to achieve quantitative assessment of multi-factor trial risk.This paper first establishes the multi-level risk index model based on AHP for the identification of trial risk indexes,which improves the analysis granularity and assessment objectivity of trial risk.Then,in order to obtain fuzzy risk measures,the risk aggregation model based on mixed input is constructed,which makes use of process information to improve the accuracy of assessment,and to realize the aggregation of multi-type trial risk indexes.Finally,the risk level determination model based on similarity algorithm is constructed to effectively eliminate the interference of human factors and abnormal data,and to realize the accurate measurement of multi-factor trial risk level.Based on the trial risk assessment model,this paper designs and implements the corresponding evaluation system,which includes risk calculation,risk management and risk early warning modules.Through integration and application of the corresponding evaluation system in the actual court management system,the trial risk management mode currently limited to time warning is effectively optimized,and the degree of refinement of trial management is further improved.
Aspect-level Sentiment Analysis of Text Based on ATT-DGRU
YIN Jiu, CHI Kai-kai, HUAN Ruo-hong
Computer Science. 2021, 48 (5): 217-224.  doi:10.11896/jsjkx.200500076
Abstract PDF(2083KB) ( 904 )   
References | Related Articles | Metrics
Aspect-level sentiment classification is to analyze the sentiment polarity in a given aspect for a given text.In the exis-ting mainstream solutions,the attention mechanism-based cyclic neural network model ignores the importance of keyword pro-ximity context information,and the CNN multilayer model is not good at capturing sentence-level long-distance dependency information.This paper proposes an aspect-level emotion classification network model based on disconnected gated recurrent units(DGRU) and attention mechanism,which is abbreviated as ATT-DGRU.The DGRU network used in this model integrates the advantages of circular neural network and CNN.It can not only capture the long-distance dependent semantic information of text,but also extract the semantic information of key phrases.Attention mechanism is used to capture the importance of each word to a specific aspect when deducing the sentiment polarity of a specific aspect,meanwhile generates an emotional weight vector,which can be visualized.Accuracies of two-class and three-class of ATT-DGRU model constructed in this paper can reach 91.11% and 87.76% respectively in ACSA task on Chinese hotel review datasets.Accuracies of two-class and three class of ATT-DGRU model can reach 77.21% and 90.06% respectively in ATSA task on SemEval2014-Restaurant dataset.
Knowledge Graph Completion Model Using Quaternion as Relational Rotation
CHEN Heng, WANG Wei-mei, LI Guan-yu, SHI Yi-ming
Computer Science. 2021, 48 (5): 225-231.  doi:10.11896/jsjkx.200300093
Abstract PDF(1515KB) ( 1155 )   
References | Related Articles | Metrics
Knowledge graph is a structured representation of real-world triples.Typically,triples are represented in the form of head entity,relationship entity and tail entity.Aiming at the data sparse problem widely existing in knowledge graph,this paper proposes a knowledge graph completion method using quaternions as relational rotation.In this paper,we model entities and relations in the expressive hyper-complex representations for link prediction.This hyper-complex embedding is used to represent entities,and relations are modelled as rotations in quaternion space.Specifically,we define each relation as a rotation from the head entity to the tail entity in the hyper-complex space,which could be used to infer and model diverse relation patterns,including symmetry/anti-symmetry,reversal and combination.In the experiment,the public datasets WN18RR and FB15K-237 are used for the related link prediction experiment.Experimental results show that on the WN18RR dataset,its mean reciprocal rank (MRR) is 4.6% higher than RotatE,and its Hit@10 is 1.7% higher than RotatE.On the FB15K-237 dataset,its MRR is 5.6% higher than RotatE,its Hit@3 is 1.4% higher than RotatE.Experiments show that the knowledge graph completion method using quaternions as relational rotation can effectively improve the prediction accuracy of triples.
Task-oriented Dialogue System and Technology Based on Deep Learning
YAO Dong, LI Zhou-jun, CHEN Shu-wei, JI Zhen, ZHANG Rui, SONG Lei, LAN Hai-bo
Computer Science. 2021, 48 (5): 232-238.  doi:10.11896/jsjkx.200600092
Abstract PDF(1660KB) ( 3115 )   
References | Related Articles | Metrics
Natural language is the crystallization of human wisdom,and interacting with computers in the form of natural language has long been expected.With the development of natural language processing technology and the rise of deep learning methods,human-computer dialogue systems have become a new research hotspot.Human-computer dialogue systems can be divided into task-oriented dialogue systems,chit-chat-oriented dialogue systems,and question-and-answer dialogue systems accor-ding to their functions.The task-oriented dialogue system is a typical man-machine dialogue system,which aims to help users complete certain specific tasks,and has very important academic significance and application value.This paper systematically illustrates the general framework of task-oriented dialogue systems in practical engineering applications,including natural language understanding,dialogue management,and natural language generation.Then,the classical deep learning and machine learning methods used in the above parts are introduced.Finally,the task of natural language understanding is empirically verified and analyzed.This paper can provide effective guidance for the construction of a task-oriented dialogue system.
Logical Reasoning Method Based on Temporal Relation Network
ZHANG Shu-nan, CAO Feng, GUO Qian, QIAN Yu-hua
Computer Science. 2021, 48 (5): 239-246.  doi:10.11896/jsjkx.201000171
Abstract PDF(3405KB) ( 722 )   
References | Related Articles | Metrics
Logical reasoning is the core of human intelligence and a challenging research topic in the field of artificial intelligence.Human IQ test is one of the common methods to measure the level of human IQ and logical reasoning ability.How to let the computer learn to have the logical reasoning ability similar to human is a very important research content,the purpose is to make the computer from a given image directly learn the logical reasoning mode without having to design a priori reasoning mode for the computer in advance.For this purpose,a new data set Fashion-IQ is proposed.Each sample of the data set contains seven input pictures and a label.The seven pictures are three question input pictures that contain one or more logics,and four option input pictures.The purpose is to let the machine learn to predict the next picture based on the logic contained in the three question input pictures,so as to select the correct option.In order to solve this problem,the temporal relationship model is proposed.For each option,the model first uses a convolutional neural network to extract the spatial features of the first three input pictures and option pictures,and then uses a relation network to combine these four spatial features in pairs.Then,it uses LSTM to extract the first three question input pictures combining the time series feature with the time series feature of this option,the time series feature and the combined space feature are combined to obtain the time series-space fusion feature.Finally,the first three input pictures and the temporal-spatial fusion features obtained from each option are further reasoned,and the softmax function is used for scoring.The option with the highest score is the correct answer.Experiments prove that the model has achieved a relatively high inference accuracy on this data set.
Named Entity Recognition in Food Field Based on BERT and Adversarial Training
DONG Zhe, SHAO Ruo-qi, CHEN Yu-liang, ZHAI Wei-feng
Computer Science. 2021, 48 (5): 247-253.  doi:10.11896/jsjkx.200800181
Abstract PDF(2236KB) ( 1299 )   
References | Related Articles | Metrics
Aiming at extracting effective entity information from unstructured corpus in the field of food safety,a named entity recognition (NER) method based on BERT (Bidirectional Encoder Representations from Transformers) and adversarial training is proposed.NER is a typical sequence labeling problem.At present,deep learning methods have been widely used in this task and have achieved remarkable results.However,there are problems such as difficulty in constructing a large number of sample sets for NER in specific fields like the food field,and inaccurate recognition of proper noun boundaries.To solve these problems,BERT is used to get the word vector,which enriches the semantic representation.To optimize the NER task,adversarial training is introduced,which not only uses the shared information obtained from task training of Chinese word segmentation (CWS) and NER,but also prevents the private information of CWS task from generating noise.The experiment is based on the corpus of two categories,which are Chinese food safety cases and People's Daily news respectively.Among them,the Chinese food safety cases data set is used to train the NER task,and the “People's Daily” news data set is used to train the CWS task.We use adversarial trainingto improve the precision of the NER task for entity recognition (including name,location,organization,food name and additive).Experimental results show that the proposed method's Precision rate,Recall rate and F1 score are 95.46%,89.50% and 92.38% respectively.Therefore,this method has a high F1 score for Chinese NER task,where the boundary of a specific domain is indistinct.
Computer Network
Performance Analysis Model of 802.11p Based Platooning Communication at Traffic Intersection
XIA Si-yang, WU Qiong, NI Yuan-zhi, WU Gui-lu, LI Zheng-quan
Computer Science. 2021, 48 (5): 254-262.  doi:10.11896/jsjkx.200700064
Abstract PDF(2659KB) ( 659 )   
References | Related Articles | Metrics
As one of the key technologies of automated driving,platooning strategy has been extensively studied and tested in practice.When platoons pass through the traffic intersection controlled by traffic light,kinetic parameters such as traffic light time,intra/inter-platoon spacing,speed,and moving direction are severely affected.At this time,since the platooning communication connectivity network is complex and variable,the driving stability of platoons is difficult to maintain,thus resulting in that the vehicles in platoons that communicate through 802.11p protocols cannot receive the complete important information within the specified delay limit,which finally leads to road safety issues.To solve this problem,considering the Enhanced Distributed Channel Access (EDCA) mechanism of the 802.11p protocol,which supports data of 4 transmission priorities to access channel,a novel communication performance analysis model for automated driving platoons under traffic intersection scenario is proposed.First,we build a connectivity network of vehicular communication at the traffic intersection and obtain the network communication performance through constructing the platoon moving model.Then,the classic Markov model that describes the 802.11p EDCA mechanism is transformed into a z-domain linear model by utilizing the method of probability generating function,and the platooning communication delay and packet delivery ratio (PDR) analysis model is deduced according to the priority differences of the 4 access categories.Finally,iterative method is used to calculate the platooning communication delay.The simulation results verify the accuracy of the analysis model,and it can be found that when passing through the intersection,the communication delay of platoons is lower than the 100ms specified by the 802.11p protocols and the packet delivery ratio is higher than 0.95,which ensures the timeliness and completeness of the platoon communication.
Diffusion Variable Tap-length Maximum Correntropy Criterion Algorithm
LIN Yun, HUANG Zhen-hang, GAO Fan
Computer Science. 2021, 48 (5): 263-269.  doi:10.11896/jsjkx.200300043
Abstract PDF(2156KB) ( 562 )   
References | Related Articles | Metrics
The fixed tap-length distributed adaptive filtering algorithm can achieve the corresponding estimation accuracy only when the tap-length of the unknown vector is assumed to be known as a prior and constant.The convergence performance of the algorithm deteriorates when the tap-length is unknown or time varying.Variable tap-length distributed adaptive filtering algorithm is an effective way to solve this problem.However,most of the distributed variable tap-length adaptive filtering algorithms use the minimum mean square error (MSE) criterion as the cost function of the tap-length,and the convergence of the algorithm is greatly affected under the impulsive noise environment.The maximum correntropy criterion is robust to impulse noise and has low computational complexity.In order to improve the estimation accuracy of the distributed variable tap-length adaptive filtering algorithm under the impulsive noise environment,the maximum correntropy criterion is used as the cost function,relevant results are substituted into the fixed tap-length diffusion maximum correntropy criterion algorithm,and thus a diffusion variable tap-length maximum correntropy criterion (DVTMCC) algorithm is proposed.By communicating with the nodes in the neighborhood,the proposed algorithm realizes the information fusion of the entire network by means of diffusion,which has advantages of a high estimation accuracy,a small calculation cost,etc.Simulation experiments compare the convergence performance of DVTMCC algorithm and other distributed variable tap-length adaptive filtering algorithms,and fix tap-length diffusion maximum correntropy criterion algorithm under the impulsive noise environment.Simulation results show that the DVTMCC algorithm can estimate the tap-length and weight vector of the unknown vector at the same time under the impulsive noise environment,and its performance is better than compared algorithms.
Deep Reinforcement Learning-based Collaborative Computation Offloading Scheme in VehicularEdge Computing
FAN Yan-fang, YUAN Shuang, CAI Ying, CHEN Ruo-yu
Computer Science. 2021, 48 (5): 270-276.  doi:10.11896/jsjkx.201000005
Abstract PDF(2248KB) ( 1380 )   
References | Related Articles | Metrics
Vehicular edge computing (VEC) is a key technology that can realize low latency and high reliability of internet of vehicles.Users offload computing tasks to mobile edge computing (MEC) servers,which can not only solve the problem of insufficient computing capability of vehicles,but also reduce the energy consumption and the latency of communication service.How-ever,the contradiction between the mobility of vehicles and the static deployment of edge servers in highway scenarios poses a challenge to the reliability of computing offloading.To solve this problem,this paper designs a collaborative deep reinforcement learning-based scheme for vehicles to adapt to the dynamic high-speed environment by combining the computing resources of MEC servers and neighboring vehicles.Simulation results show that compared with the scheme without vehicle collaboration,this scheme can reduce the delay and the failure rate of offloading.
Channel Assignment Algorithm Based on Particle Swarm Optimization in Emergency Communication Networks
LIU Wei, LI Dong-kun, XU Chang, TIAN Zhao, SHE Wei
Computer Science. 2021, 48 (5): 277-282.  doi:10.11896/jsjkx.200400042
Abstract PDF(1842KB) ( 757 )   
References | Related Articles | Metrics
How to quickly and effectively meet the rapidly increasing network demand and ensure the quality of network transmission is the problem in emergency communications that needs to be solved urgently.The wireless mesh network is the choice of a new generation of post-disaster emergency communication network architecture.This paper proposes a particle swarm optimization-based channel allocation algorithm that considers the impact of different links on the overall network performance under the premise of reducing global network interference.The priority of channel allocation is determined by the number of adjacent links.In the process of iteration,the channel separation is used to subdivide the degree of interference between different channels as the criterion for optimization.The experimental results show that the proposed algorithm can significantly reduce networkinterfe-rence and ensure network performance.Compared with the traditional particle swarm allocation algorithm,its optimization speed and performance are significantly improved in the multi-node network environment.
Dynamic Allocation Mechanism of Preamble Resources Under H2H and M2M Coexistence Scenarios
WANG Cong, WEI Cheng-qiang, LI Ning, MA Wen-feng, TIAN Hui
Computer Science. 2021, 48 (5): 283-288.  doi:10.11896/jsjkx.200300019
Abstract PDF(2515KB) ( 614 )   
References | Related Articles | Metrics
As massive M2M devices are connected to the network,the network performance declines sharply.At the same time,due to limited preamble resources,the access success probability of H2H users is severely reduced.To solve this problem,this paper proposes a dynamic allocation mechanism of preamble resources in hybrid H2H and M2M scenarios.In this method,on the condition that the H2H average access delay meets requirements,the amount of preamble resources allocated to M2M devices is dynamically adjusted.Then,according to the amount of allocated preamble resources,the number of M2M devices competing in each random-access opportunity is dynamically adjusted to maximize the access efficiency of M2M devices.Through the simulation of the success probability of M2M devices and the average H2H access delay,experimental results show that this method significantly improves the success probability of M2M devices when the average H2H access delay is low,compared with the fixed resource allocation mechanism.
Chaotic Prediction Model of Network Traffic for Massive Data
XIANG Chang-sheng, CHEN Zhi-gang
Computer Science. 2021, 48 (5): 289-293.  doi:10.11896/jsjkx.200400056
Abstract PDF(1941KB) ( 629 )   
References | Related Articles | Metrics
Aiming at the chaotic and massive characteristics of network traffic,in order to make up for the shortcomings of network traffic prediction model to obtain better network traffic prediction results,a chaotic network traffic prediction model for massive data is proposed.First,wavelet analysis is used to deal with the original network traffic time series in multi-scale to obtain network traffic components with different characteristics.Then,the chaotic characteristics of network traffic components are analyzed and reconstructed respectively.The extreme learning machine in machine learning algorithm is used to model and predict.Finally,wavelet analysis is used to overlay the prediction results of network traffic components to get the original network traffic data prediction value,and the network traffic prediction simulation experiment is carried out.Experimental results show that,compared with other network traffic prediction models,the network traffic prediction accuracy of the proposed model is more than 90%,and the network traffic prediction results are more stable.It is an effective tool for network traffic modeling and prediction.
Information Security
Attack Path Analysis Method Based on Absorbing Markov Chain
ZHANG Kai, LIU Jing-ju
Computer Science. 2021, 48 (5): 294-300.  doi:10.11896/jsjkx.200700108
Abstract PDF(2539KB) ( 692 )   
References | Related Articles | Metrics
The analysis of network attack path from the perspective of attackers is of great significance to guide network security defense.The existing analysis methods based on absorbing Markov chain have some problems,such as incomplete consideration of state transition and unreasonable calculation of state transition probability.In order to solve these problems,this paper proposes an attack path analysis method based on absorbing Markov chain.Based on the generation of attack graph and the exploitability score of vulnerability,the situation that the failure state transition of non-absorbing nodes will be fully considered.In order to map the attack graph to the absorbing Markov chain model,this paper proposes a new method to calculate the state transition probability.Then,by using the properties of the state transition probability matrix of the absorbing Markov chain,it calculates the threat ranking of the nodes in the attack path and the expected length of the attack path.Then,the application feasibility of absorbing Markov chain with multi absorbing states is discussed.The results of the experiment show that the proposed method can effectively calculate the node threat ranking and path length expectation.Through comparative analysis,this method is more in line with the actual situation of network attack and defense than the existing methods.
Team Cooperative Attack Planning Based on Multi-agent Joint Decision
ZHOU Tian-yang, ZENG Zi-yi, ZANG Yi-chao, WANG Qing-xian
Computer Science. 2021, 48 (5): 301-307.  doi:10.11896/jsjkx.200800174
Abstract PDF(2272KB) ( 981 )   
References | Related Articles | Metrics
Automated penetration testing can greatly reduce the cost of penetration testing by automating the process of manually finding possible attack paths.Existing methods mainly use a single agent to perform attack tasks,which leads to long execution of attack actions and low penetration efficiency.If multi-agent cooperative attack is considered,the state space scale of planning problem will grow exponentially due to the multi-dimensional local state of each agent.Aiming at the above problems,a team cooperative attack planning method based on multi-agent jointdecision is proposed.Firstly,the multi-agent cooperative attack path planning problem is transformed into the attack target assignment problem under the jointdecision constraints,and themulti-agent centralized decision-making mode is established.Secondly,the joint decision vector matrix JDVM is used to calculate the penetration attack reward based on the CDSO-CAP model,and the greedy strategy is used to search the optimal target of attack.The experimental results show that compared with the single agent planning method,the proposed method has similar algorithm convergence with shorter execution rounds.Thus it is more suitable for rapid attack planning in multi-target network scenarios.
Resilient Distributed State Estimation Algorithm
GAO Feng-yue, WANG Yan, ZHU Tie-lan
Computer Science. 2021, 48 (5): 308-312.  doi:10.11896/jsjkx.200300117
Abstract PDF(1912KB) ( 786 )   
References | Related Articles | Metrics
In order to improve the immunity of multi-agent system against attack,resilient distributed state estimation under measurement attacks is studied.Each agent makes successive local linear measurements of the system state.The local measurement models are heterogeneous across agents and may be locally unobservable for the system state.An adversary compromises some of the measurement streams and changes their values arbitrarily.The agents' goal is to cooperate with their local measurements and estimate the value of the system state correctly.The challenge of this problem is how to design an algorithm to estimate the real system state without distinguishing the real measurements from the measurements of malicious agents.In order to solve this problem,an adaptive distributed maximum a posteriori probability estimation algorithm is designed.As long as the number of compromised measurement streams is lower than a particular bound,all of the agents' local estimates,including malicious agents' local estimates,can converge to the true system state.Firstly,a centralized maximum a posteriori (MAP) estimation method is proposed based on Kalman filter.Combining a centralized MAP estimation with distributed consensus protocol,a distributed MAP estimation method is derived.Then,considering the measurement attack and analyzing the consistency of distributed estimates,a resilient distributed MAP estimation method is designed by exploiting the saturating adaptive gain,which gives a small gain if the deviation from the practical measurement resulting from the attacks is too large.At last,Numerical simulations are provided to evaluate the effectiveness of the proposed algorithm against measurement attacks.
Attribute Access Control Based on Dynamic User Trust in Cloud Computing
PAN Rui-jie, WANG Gao-cai, HUANG Heng-yi
Computer Science. 2021, 48 (5): 313-319.  doi:10.11896/jsjkx.200400013
Abstract PDF(1949KB) ( 812 )   
References | Related Articles | Metrics
In order to facilitate the management of resources in the cloud,the cloud computing environment is usually divided into logically independent security management domains,but there is a hidden danger in the loss of resources' physical boundary protection.Access control is one of the key technologies to solve this security problem.Aiming at the characteristic of multiple domains of cloud computing environment,this paper proposes an access control model (CT-ABAC) based on dynamic user trust to reduce the impact of malicious recommendations in the security domain and reduce the number of malicious users' visits.In the CT-ABAC model,an access request consists of subject attributes,object attributes,permission attributes,environment attributes,and user trust attributes.A dynamic fine-grained authorization mechanism is used to deny or allow this access based on the user'saccess request attribute set.At the same time,this model extends the attribute of user trust,and considers the impact of time,similarity between security domains,and penalty mechanisms on this attribute.Simulation results show that the proposed model can effectively reduce the malicious access of users and improve the success rate of trusted users.
Parallel Multi-keyword Top-k Search Scheme over Encrypted Data in Hybrid Clouds
JI Yan, DAI Hua, JIANG Ying-ying, YANG Geng, Yi Xun
Computer Science. 2021, 48 (5): 320-327.  doi:10.11896/jsjkx.200300160
Abstract PDF(2625KB) ( 670 )   
References | Related Articles | Metrics
With the rapid development of cloud computing services,more and more individuals and enterprises prefer to outsource and manage their data or computation to clouds.In order to protect the privacy of outsourced data,encryption before outsourcing is a commonly measure.However,it is a challenge to perform searches over encrypted data.In this paper,we propose a parallel privacy-preserving multi-keyword Top-k search scheme over encrypted data in hybrid clouds.The scheme can protect the privacy of outsourced data and support multi-keyword search over encrypted data,which performs vectorization on documents and keyword groups and introduces the symmetric encryption and the homomorphic matrix encryption.In addition,the scheme adopts the MapReduce model to perform parallel searches in the public clouds and the private clouds.Thus,parallel searches over the large scale encrypted data are achieved.The security analysis and the performance evaluation show that the proposed scheme is a privacy-preserving multi-keyword Top-k search scheme and outperforms the existing scheme in terms of search efficiency.
Authenticable Encrypted Secure Communication Based on Adversarial Network
WU Shao-qian, LI Xi-ming
Computer Science. 2021, 48 (5): 328-333.  doi:10.11896/jsjkx.200300177
Abstract PDF(2094KB) ( 762 )   
References | Related Articles | Metrics
Since GANs(Generative Adversarial Networks) has been put forward,it has been widely used in various fields,and its application in the field of information security is getting more and more deeply.However,using GANs to realize secure communication under public key cryptosystem has not been reported publicly.Therefore,in view of the adversarial nature of both communication sides and their adversary,this paper proposes an adversarial learning mechanism of GANs.In the public key cryptosystems scenarios,the key generator,encryption and decryption of both communication sides,and the decipher process of adversary are regarded as neural networks,then we use the certification confidentiality to strengthen public-private key linkage.Afterwards,by using the adversarial learning mechanism to train both communication sides and their adversary,we realize the authenticable encrypted secure communication (AEC-AN) between both communication sides on the open channel.In the experiment,4 keys with lengths of 16 bit,32 bit,64 bit and 128 bit have been used for training.The experiment result shows that Bob's accuracy rate is between 91%~94%,and Eve's error rate is between 43%~57%,which is close to the probability of Eve's random guess,thus proving that the proposed mechanism of GANs achieves the secure communication between both communication sides under the environment of adversary eavesdropping.
Source-location Privacy Protection Scheme Based on Target Decision in WSN
GUO Rui, LU Tian-liang, DU Yan-hui
Computer Science. 2021, 48 (5): 334-340.  doi:10.11896/jsjkx.200400099
Abstract PDF(2328KB) ( 793 )   
References | Related Articles | Metrics
Aiming at the problem that the existing schemes of Wireless Sensor Network(WSN) source-location privacy protection based on phantom source can not effectively balance the contradiction among source location privacy security,network life cycle and transmission delay,a phantom source separate path routing scheme (PSSR) based on target decision is proposed.In PSSR scheme,the phantom node location is determined by random walk of segmented fixed image,which ensures that the phantom source is far enough from the real source visible area,and at the same time realizes the diversity of phantom source location,which increases the difficulty of attacker locating the source location.In addition,by considering the energy consumption of the node,the remaining energy and the distance from the node to the base station,the forwarding node is selected to realize the construction of low probability repeated and decentralized routing,effectively balancing the contradiction among the source location privacy security,network life cycle and transmission delay.Compared with EPUSBRF,PRLA and MPRP,PSSR can not only enhance the source location privacy security,but also effectively prolong the network lifetime and reduce the transmission delay.
渝ICP备16013121号-4
Add:18# Honghu West Rd., Yubei ,Chongqing,China    Post Code: 401121
Tel: 023-63500828/023-67039612/023-67039623
E-mail: jsjkx12@163.com
Powered by Beijing Magtech Co., Ltd.