Started in January,1974(Monthly)
Supervised and Sponsored by Chongqing Southwest Information Co., Ltd.
ISSN 1002-137X
CN 50-1075/TP
CODEN JKIEBK
Editors
Current Issue
Volume 48 Issue 10, 15 October 2021
  
Artificial Intelligence
Survey of Reinforcement Learning Based Recommender Systems
YU Li, DU Qi-han, YUE Bo-yan, XIANG Jun-yao, XU Guan-yu, LENG You-fang
Computer Science. 2021, 48 (10): 1-18.  doi:10.11896/jsjkx.210200085
Abstract PDF(2455KB) ( 3289 )   
References | Related Articles | Metrics
Recommender systems are devoted to find and automatically recommend valuable information and services for users from massive data,which can effectively solve the information overload problem,and become an important information technology in the era of big data.However,the problems of data sparsity,cold start,and interpretability are still the key technical difficulties that limit the wide application of the recommender systems.Reinforcement learning is an interactive learning technique,which can dynamically model user preferences by interacting with users and obtaining feedback to capture their interest drift in real time,and can better solve the classical key issues faced by traditional recommender systems.Nowadays,reinforcement lear-ning has become a hot research topic in the field of recommendation systems.From the perspective of survey,this paper first analyzes the improvement ideas of reinforcement learning for recommender systems based on a brief review of recommender systems and reinforcement learning.Then,the paper makes a general overview and summary of reinforcement learning based recommender systems in recent years,and further summarizes the research situation of traditional reinforcement learning based recommendation and deep reinforcement learning based recommendation respectively.Furthermore,the paper summarizes the frontiers of reinforcement learning based recommendation research topic in recent years and its application.Finally,the future development trend and application of reinforcement learning in recommender systems are analyzed.
Overview of Global Path Planning Algorithms for Mobile Robots
WANG Zi-qiang, HU Xiao-guang, LI Xiao-xiao, DU Zhuo-qun
Computer Science. 2021, 48 (10): 19-29.  doi:10.11896/jsjkx.200700114
Abstract PDF(1835KB) ( 3453 )   
References | Related Articles | Metrics
Global path planning is the key technology of mobile robot outdoor work,and global path planning algorithm is mainly used in geographical scenarios to predict the outdoor environment.In the face of the complex outdoor environment,the robot optimizes the algorithm to improve real-time obstacle avoidance of robot path planning,path smoothness,planning effectiveness,which has become the core content in the research of global path planning algorithm.Depending on the degree of intelligent algorithm,the global path planning algorithm for mobile robot is divided into traditional global path planning algorithm and the bionic intelligent global path planning algorithm.Then,this paper further expounds the current practical multi-objective path planning algorithm,introduces several typical optimizations of each algorithm,and summarizes and analyzes the advantages and disadvantages of the improved algorithm.Finally,this paper discusses the future development trend of global path algorithm,and points out four aspects of future research and development,which are optimizing the performance of the conventional algorithm for path planning,various existing algorithms advantage integration,complex environment dynamic obstacle avoidance and improving map representation methods adapting to diverse environment.
Monte Carlo Tree Search for High-dimensional Continuous Control Space
LIU Tian-xing, LI Wei, XU Zheng, ZHANG Li-hua, QI Xiao-ya, GAN Zhong-xue
Computer Science. 2021, 48 (10): 30-36.  doi:10.11896/jsjkx.201000129
Abstract PDF(2138KB) ( 1550 )   
References | Related Articles | Metrics
Monte Carlo tree search (MCTS) has gained great success in low discrete control tasks.However,there are many tasks in real life that require selecting action sequentially in continuous action space.Kernel regression UCT (KR-UCT) is a successful attempt in low-dimensional continuous action space by using a pre-defined kernel function to exploit the similarity of different continuous actions.However,KR-UCT gets a poor performance when it comes to high-dimensional continuous action space,because KR-UCT does not use the interacting information between agent and the environment.And when it interacts with the environment,KR-UCT needs to perform a lot of simulations at each step to find the best action.In order to solve this problem,this paper proposes a method named kernel regression UCT with policy-value network (KRPV).The proposed method can filter out more representative actions from action space to perform MCTS and generalize the information between different states to pruning MCTS.The proposed method has been evaluated by four continuous control tasks of the OpenAI gym.The experimental results show that KRPV outperforms KR-UCT in all tested continuous control tasks.Especially for the six-dimensional HalfCheetah-v2 task,the rewards gained by KRPV are six-timesof that of KR-UCT.
Deep Deterministic Policy Gradient with Episode Experience Replay
ZHANG Jian-hang, LIU Quan
Computer Science. 2021, 48 (10): 37-43.  doi:10.11896/jsjkx.200900208
Abstract PDF(2351KB) ( 1030 )   
References | Related Articles | Metrics
The research on continuous control in reinforcement learning has been a hot topic in recent years.The deep deterministic policy gradient (DDPG) algorithm performs well in continuous control tasks.DDPG algorithm uses experience replay mechanism to train the network model,and in order to further improve the efficiency of experience replay mechanism in the DDPG algorithm,the cumulative reward is used as the transition classification basis,a deep deterministic policy gradient with episodic experience replay (EER-DDPG) algorithm is proposed.First of all,the transitions are stored in the unit of episode,and two replay buffersare introduced respectively to classify the transitions according to the cumulative reward.Then,the quality of policy can be improved in network model training period by random sampling of the episodes with large cumulative reward.In the continuous control tasks,this algorithm is verified by experiments,and compared with DDPG algorithm,trust region policy optimization (TRPO) algorithm and proximal policy optimization (PPO) algorithm.The experimental results show that EER-DDPG algorithm has better performance.
Graph Based Collaborative Extraction Method for Keywords and Summary from Documents
MAO Xiang-ke, HUANG Shao-bin, YU Qin-yong
Computer Science. 2021, 48 (10): 44-50.  doi:10.11896/jsjkx.200900082
Abstract PDF(1767KB) ( 766 )   
References | Related Articles | Metrics
The purpose of keywords extraction and summary extraction is to select key content from the original document to express the main meaning of the original document.The evaluation of keywords and summarization quality mainly depends on whether it can cover the main topics of the document.In the existing methods of keywords extraction and summary extraction based on graph models,it rarely involves the task of keywords extraction and summary extraction collaboratively.The article proposes a method based on a graph model for simultaneous keywords extraction and summary extraction.The method first uses the six relationships among words,topics,and sentences in the document,including words-words,topics-topics,sentences-sentences,words-topics,topics-sentences,words-sentences,to construct the graph;then uses the statistical characteristics of the words and sentences in the document to evaluate the prior importance of each vertex in the graph;next,it uses an iterative way to score words and sentences;finally,we get the final keywords and summary based on the scores of words and sentences.In order to verify the effectiveness of the proposed method,keywords extraction and summary extraction experiments are carried out on Chinese and English datasets.It is found that the proposed method achievs good results in both keywords extraction and summary extraction tasks.
Fusion Vectorized Representation Learning of Multi-source Heterogeneous User-generated Contents
JI Nan-xun, SUN Xiao-yan, LI Zhen-qi
Computer Science. 2021, 48 (10): 51-58.  doi:10.11896/jsjkx.200900194
Abstract PDF(3514KB) ( 783 )   
References | Related Articles | Metrics
With the development of mobile networks and APPs,user generated contents (UGC) containing multi-source heterogeneous data such as evaluations,markings,scoring,images and videos are greatly valuable information for improving the quality of personalized services.The representation learning of fusion and vectorization on the multi-source heterogeneous UGC is the most critical issue for the successful application.Motivated by this,we propose a representation learning method for effectively fusing and vectorizing the comments and image data.We utilize the Doc2vec and LDA models to sufficiently extract the features of the multi-source comments.The images correlated with the comments are represented with deep convolutional network.A hybrid vectorized representation learning for fusing comments and a convolution strategy for integrating images and comments are presented.The feasibility and effectiveness of the proposed method is demonstrated by applying it to typical Amazon public data sets with heterogeneous UGC,in which the vectorized multi-source heterogeneous UGC is taken as the representation of each product and the classification accuracy of the products are compared.
Scientific Paper Summarization Using Word-Section Association
FU Ying, WANG Hong-ling, WANG Zhong-qing
Computer Science. 2021, 48 (10): 59-66.  doi:10.11896/jsjkx.200900180
Abstract PDF(3789KB) ( 557 )   
References | Related Articles | Metrics
With the development of science and technology,people need to access a large number of scientific and technological information quickly,and scientific paper is one of the main ways to carry scientific and technological information.As an important part of scientific paper,abstract is an effective tool for readers to retrieve literature.Therefore,the quality of abstract affects the retrieval rate of paper directly.However,due to the lack of writing experience,the quality of abstracts written by many authors is not high.Automatic generation of summary for scientific paper can help the author grasp the important content of paper more effectively,so as to write high-quality abstract.At the same time,the automatically generated abstract can also control the number of words in the abstract,which can bring more content to readers and help them understand the paper better.Generating automa-tic summarization for scientific paper can help author write abstract faster,which is one of the research contents in automatic summarization.Compared with common news document,scientific paper has the characteristics of strong structure and clear logical relationship.As far as the mainstream abstractive summarization such as encoder-decoder model is concerned,it mainly consi-ders the serialized information in the document,and rarely explores the text structure information in the document.For this reason,according to the characteristics in scientific papers,this paper proposes an automatic summarization model based on the hie-rarchical structure of “word-section-document”,which uses the association between word and section to enhance the level of text structure and the interaction between levels,so as to screen out the key information in scientific paper.In addition,a context gate unit is extended to update the optimized context vector,thus capturing context information more comprehensively.The experimental results show that the proposed model can effectively improve the performance of the generated summarization in the ROUGE evaluation method.
Multi-subgroup Particle Swarm Optimization Algorithm with Game Probability Selection
TIAN Meng-dan, LIANG Xiao-lei, FU Xiu-wen, SUN Yuan, LI Zhang-hong
Computer Science. 2021, 48 (10): 67-76.  doi:10.11896/jsjkx.200800128
Abstract PDF(5821KB) ( 607 )   
References | Related Articles | Metrics
Aimed at solving the defects of premature and easy being trapped into the local optimum of particle swarm optimization (PSO),a new algorithm is proposed with considering the species group structure,multi-mode learning and individual game,which was named as multi-subgroup particle swarm optimization algorithm with game theory (MPSOGT).The proposed algorithm constructs a dynamic multi-subgroup structure and introduces different learning strategies to form a multi-source learning strategy with heterogeneous multiple subgroups.Then the evolutionary game theory is introduced into the process of population sear-ching.According to a dynamic payoff matrix and a dynamic rooted probability based on the game theory,each individual enters into a suitable subgroup randomly to enhance its searching ability.Based on twelve benchmark functions,combined experiments are carried out for subgroup size L.The results show that the population fitness and median have obvious advantages when the value of L is N/2 or N/3.Compared with seven similar algorithms under a set of twelve benchmark functions test with different scales,the results show that the performance of the improved algorithm is superior to the comparison algorithm in terms of optimal va-lue,solution stability and convergence characteristics.It is indicated that the proposed multi-source learning and game probability selecting strategies can effectively improve the performance of the PSO algorithm.
Medical Entity Relation Extraction Based on Deep Neural Network and Self-attention Mechanism
ZHANG Shi-hao, DU Sheng-dong, JIA Zhen, LI Tian-rui
Computer Science. 2021, 48 (10): 77-84.  doi:10.11896/jsjkx.210300271
Abstract PDF(1514KB) ( 895 )   
References | Related Articles | Metrics
With the advancement of medical informatization,a large amount of unstructured text data has been accumulated in the medical field.How to mine valuable information from these medical texts is a research hotspot in the field of medical profession and natural language processing.With the development of deep learning,deep neural network is gradually applied to relation extraction task,and “recurrent+CNN” network framework has become the mainstream model in medical entity relation extraction task.However,due to the problems of high entity density and the cross-connection of relationships between entities in medical texts,the “recurrent+CNN” network framework cannot deeply mine the semantic features of medical texts.Based on the “recurrent+CNN” network framework,this paper proposes a Chinese medical entity relation extraction model with multi-channel self-attention mechanism.It includes that BLSTM is used to capture the context information of text sentences,a multi-channel self-attention mechanism is used to mine the global semantic features of sentences,and CNN is used to capture the local phrase features of sentences.The effectiveness of the model is verified by experiments on Chinese medical text dataset.The precision,recall and F1 value of the model are improved compared with the mainstream models.
Chinese Implicit Discourse Relation Recognition Based on Data Augmentation
WANG Ti-shuang, LI Pei-feng, ZHU Qiao-ming
Computer Science. 2021, 48 (10): 85-90.  doi:10.11896/jsjkx.200800115
Abstract PDF(1439KB) ( 841 )   
References | Related Articles | Metrics
Due to the lack of connectives,implicit discourse relation recognition is a challenging task,especially in Chinese.This paper proposes a method for Chinese implicit discourse relation recognition,which expands the training data by combining active learning and multi-task learning method.This method aims to reduce the noise as much as possible when it expands the training data set.Firstly,the active learning is used to select some explicit data through the classification uncertainty based on BERT,and then the connectives in the explicit data are removed and regarded as pseudo-implicit training data.Finally,a multi task learning method is used to boost implicit discourse relation recognition by using the pseudo-implicit training data.Experimental results on Chinese discourse treebank (CDTB) show that our method improves the macro-average F1 and micro-average F1 scores,compared with the baselines.
Entity Recognition Fusing BERT and Memory Networks
CHEN De, SONG Hua-zhu, ZHANG Juan, ZHOU Hong-lin
Computer Science. 2021, 48 (10): 91-97.  doi:10.11896/jsjkx.200900015
Abstract PDF(2056KB) ( 889 )   
References | Related Articles | Metrics
Entity recognition is a sub task of information extraction.The traditional entity recognition model is used to identify entities of personnel,organization,location and name.In the real world,more types of entities must be considered,and fine-grained entity recognition is needed.At the same time,traditional entity recognition models such as BiGRU cannot make full use of the global features in a wider range.This paper presents an entity recognition model based on memory network and BERT.The pre-training language model of BERT is used for better semantic representation,and the memory network module can memorize a wider range of features.The results of entity recognition for cement clinker production corpus data show that this method can re-cognize entities and has some advantages over other traditional models.In order to further verify the model in this paper,experiments are carried out on the CLUENER2020 dataset.The results show that the optimization based on BiGRU-CRF model using BERT and memory network module can improve the effect of entity recognition.
Optimal Granulation Selection Method Based on Multi-granulation Rough Intuitionistic Hesitant Fuzzy Sets
XUE Zhan-ao, SUN Bing-xin, HOU Hao-dong, JING Meng-meng
Computer Science. 2021, 48 (10): 98-106.  doi:10.11896/jsjkx.200800074
Abstract PDF(1373KB) ( 492 )   
References | Related Articles | Metrics
In order to obtain the optimal granulations after reduction from the intuitionistic hesitant fuzzy decision information system with multiple attributes,this paper deals with the uncertain information in this system from the perspective of multi-gra-nulation rough sets,and studies optimal granulation selection method based on multi-granulation rough intuitionistic hesitant fuzzy sets.Firstly,on the basis of intuitionistic hesitant fuzzy sets,attribute information is introduced,and the concept of rough intui-tionistic hesitant fuzzy sets is given.Then four upper and lower approximation models of optimistic and pessimistic multi-granulation rough intuitionistic hesitant fuzzy sets are proposed,and the related properties are discussed.Secondly,mainly based on the lower approximation of the pessimistic multi-granulation rough intuitionistic hesitant fuzzy set,this paper defines the granu-lation quality similarity degree and internal/external granulation importance degree,and the related algorithm of optimal granulation selection is designed.Finally,through the wine evaluation case,optimal granularities are calculated based on the four cases of optimistic and pessimistic multi-granulation rough intuitionistic hesitant fuzzy set's upper and lower approximation,then analyzes results.It is verified that algorithms are effective for the reduction of intuitionistic hesitant fuzzy decision information system.
Multimodal Representation Learning for Alzheimer's Disease Diagnosis
FAN Lian-xi, LIU Yan-bei, WANG Wen, GENG Lei, WU Jun, ZHANG Fang, XIAO Zhi-tao
Computer Science. 2021, 48 (10): 107-113.  doi:10.11896/jsjkx.200900178
Abstract PDF(2407KB) ( 822 )   
References | Related Articles | Metrics
Alzheimer's disease (AD) is a complex neurodegenerative disease involving a variety of pathogenic factors.So far,the cause of Alzheimer's disease is not clear,the course of the disease is irreversible,and there is no cure.Its early diagnosis and treatment have always been the focus of attention.The neuroimaging data of subjects has an important auxiliary role in the diagnosis of this disease,and the combination of multimodal data can further improve the diagnostic effect.At present,the multimodal data representation learning of the disease has gradually become an emerging research field,which has attracted wide attention from researchers.An autoencoder based multimodal representation learning method for Alzheimer's disease diagnosis is proposed.Firstly,the multimodal data are initially fused to obtain the primary common representation.Then,it is input into the autoencoder network to learn the final common representation in latent space.Finally,the common representation in latent space is classified to obtain the disease result.The proposed method,which achieves the best diagnostic results compared with comparison algorithms,has an accuracy of 88.9% in the classification of AD and healthy subjects in the ADNI dataset.Extensive experimental results verify its effectiveness.
miRNA-disease Association Prediction Model Based on Stacked Autoencoder
LIU Dan, ZHAO Sen, YAN Zhi-liang, ZHAO Jing, WANG Hui-qing
Computer Science. 2021, 48 (10): 114-120.  doi:10.11896/jsjkx.200900169
Abstract PDF(2967KB) ( 760 )   
References | Related Articles | Metrics
As a group of small non-coding RNA,the abnormal regulation of miRNA is closely related to the occurrence and deve-lopment of human diseases.The study on the associations between miRNA and disease is important for understanding the pathogenic mechanism of human diseases.Machine learning methods are widely used to predict miRNA-disease associations.However,existing methods only consider the information of miRNA and disease similarity networks,ignoring the topology structure of the similarity networks.Therefore,SAEMDA model based on stacked autoencoder is proposed in this paper,it gets the topological structure features of miRNA and disease similarity networks by restart random walk,obtains the abstract low dimensional features of miRNA and disease by stacked autoencoder,and the low dimensional features are input into deep neural network for miRNA-disease associations prediction.SAEMDA model has achieved great results in 5-fold cross-validation,and it has been validated in cases of colon cancer and lung cancer additionally.As for colon cancer,45 of the top 50 miRNA-disease associations predicted by this model are verified in the database;and in the cases of lung cancer,all the top 50 miRNAs are verified in the database.
Study on ECG Signal Recognition and Classification Based on U-Net++
YANG Chun-de, JIA Zhu, LI Xin-wei
Computer Science. 2021, 48 (10): 121-126.  doi:10.11896/jsjkx.200700103
Abstract PDF(3062KB) ( 1101 )   
References | Related Articles | Metrics
It is difficult to explore efficient,fast and accurate ECG signal recognition and classification algorithm.The classification of ECG fragments is more suitable for clinical application.Based on this,the improved generation countermeasure network (DCGAN) is used for data expansion,and the optimized one-dimensional U-Net++ is used for fragment signal recognition of arrhythmia.ECG fragments from 1 200 sampling points in MIT-BIH database are continuously intercepted as the experimental data set,and the type that appears the most times of beats in each fragment recording center is used as the label of the whole record.Then the DCGAN,which uses optimized one-dimensional U-Net++ as generator,is used to realize partial data expansion to solve the problem of data imbalance.Under the condition that the original ECG signals are not preprocessed and the generated extended data are used to complete the wavelet threshold denoising,the accuracy of the optimized one-dimensional U-Net++ model for normal,ventricular premature beat,left bundle branch block,right bundle branch block four kind of different type can reach 98.10% for the training sets.The precision ratio,recall ratio and F1 score of the test set have good results.Under the same experimental data set,the accuracy of U-Net++ model is 1.05% higher than that of U-Net model.Under the same experimental parameters,compared with under sampling data,the accuracy of the experimental model of the data set expanded by DCGAN is improved by 0.85%.
Drug Target Interaction Prediction Method Based on Graph Convolutional Neural Network
GAO Chuang, LI Jian-hua, JI Xiu-yi, ZHU Cheng-long, LI Shi-liang, LI Hong-lin
Computer Science. 2021, 48 (10): 127-134.  doi:10.11896/jsjkx.200700068
Abstract PDF(2247KB) ( 1214 )   
References | Related Articles | Metrics
Drug-target interaction prediction plays an important role in drug discovery and repositioning.However,existing prediction methods have the problem of insufficient predictive performance while processing data with highly unbalance positive and negative samples.Therefore,a novel computational method based on graph convolutional neural network(GCN) for predicting drug-target interactions is proposed.In this method,a heterogeneous information network is constructed,which integrates diverse drug-related information and target-related information.From the heterogeneous information network,low-dimensional vector representation of features,which accurately explains the topological properties of individual and neighborhood feature information,is learned by using GCN and then prediction is made based on these representations via a vector space projection scheme.The AUPR(Area Under the Precision-Recall Curve) values of the proposed method outperforms other four existing methods in the prediction of drug-target interaction on both DrugBank_FDA and Yammanishi_08 datasets,and it preforms well on bigger datasets.The experimental results indicate that the proposed method improves the prediction performance of drug-target interaction on datasets with highly unbalanced samples.Furthermore,we validate novel(unknown) drug-target interactions which are predicted by GCN in biomedical databases.
Control Application of Wolf Group Optimization Convolutional Neural Network in Ship Virtual Manufacturing
XIAO Shi-long, WU Di, TANG Chao-chen, SHEN Xian-hao, ZHANG De-yu
Computer Science. 2021, 48 (10): 135-139.  doi:10.11896/jsjkx.200900183
Abstract PDF(1691KB) ( 486 )   
References | Related Articles | Metrics
In order to optimize the control strategy of virtual industrial manufacturing,the convolution neural network algorithm based on wolf swarm optimization is used to study the control of virtual industrial manufacturing.Firstly,according to the task and resource data of virtual industrial manufacturing,the task resource list is established,and the task resource list is sparse combined with the unit matrix to form the virtual manufacturing cell.Then,the convolution neural network virtual manufacturing control model is established,and the weight and offset are optimized by using wolf swarm algorithm.Finally,the average manufacturing time of all tasks is taken as the objective function and the manufacturing unit is trained and optimized.The virtual manu-facturing experiment of marine main engine shows that compared with the common control algorithm,the convolution neural network algorithm optimized by wolves can obtain better average manufacturing time by setting the pool size of convolution kernel reasonably.
Database & Big Data & Data Science
Conversion Method from Relational Database to Graph Database
E Hai-hong, HAN Peng-hao, SONG Mei-na
Computer Science. 2021, 48 (10): 140-144.  doi:10.11896/jsjkx.201100073
Abstract PDF(1410KB) ( 1656 )   
References | Related Articles | Metrics
Due to the differences between the storage mode of relational database and graph database,during the process of transforming data in relational database to graph database,it is necessary to solve the main problems of edge definition,vertex uniqueness and retention of original database constraint information.To solve the above problems,a method of transforming relational database to graph database is proposed.Firstly,by customizing the existing primary key,combined with the uniqueness of the table name,the problem of ensuring the uniqueness of the vertex is solved;through different configuration schemes,the constraint information of the original relational database can be maximized.Then,the edge definition method based on configuration and intermediate table (EDCIT) method is proposed,it provides different edge mapping solutions for multiple types of databases and solves the definition of edges during the transformation.Finally,through experiments on multiple data sets,and using Gremlin statement to test the transformed data,it verifies the integrity and reliability of the transformed data.
Mapping Method from Object-relational Database to RDF(S)
LU Jia-wen, YAN Li
Computer Science. 2021, 48 (10): 145-151.  doi:10.11896/jsjkx.200800006
Abstract PDF(2604KB) ( 792 )   
References | Related Articles | Metrics
With the development of intelligent information technology,knowledge graph has been widely used in intelligent search and other research area.The information in the knowledge map is generally represented by the data model of RDF(S).The construction of knowledge graph needs to extract information from different data sources and database is an important data source that cannot be ignored.Nowadays,object-relational databases are widely used and contain rich semantic information,but research on constructing RDF(S) from object-relational databases is few.This paper puts forward formal definitions of object-relational databases and RDF(S) data and proposes mapping rules for constructing RDF(S) data from object-relational databases.The mapping rules not only consider the object-oriented semantics of the database,but also consider constraints,which can fully extract semantic information contained in the database.Finally,a mapping tool named ORDB2RDF is implemented to verify the correctness of the mapping rules and the semantic integrity of the mapping results.
Top-k Densest Subgraphs Search in Temporal Graphs
MU Cong-cong, WANG Yi-shu, YUAN Ye, QIAO Bai-you, MA Yu-liang
Computer Science. 2021, 48 (10): 152-159.  doi:10.11896/jsjkx.201100005
Abstract PDF(2392KB) ( 745 )   
References | Related Articles | Metrics
The dense subgraph search problem is one of the most important graph analysis problems.It is widely used in many fields,such as the social user correlation analysis in social networks,the community discovery in the Web,etc.However,the current research focuses on searching dense subgraphs on static graphs.In practical application,temporal information has an important impact on the query of dense subgraphs,which makes the topology structure of graphs change constantly with time sequences,and the amount of information contained also increases dramatically.Therefore,the existing searching methods for static graphs are no longer applicable to temporal graphs.Hence,it is still a challenge to search dense subgraphs efficiently on a temporal graph.In order to solve the above challenge,this paper first defines the Top-k densest temporal subgraphs searching problems in a standardized way.Then,to address the above challenge,this paper proposes an approximate searching algorithm DTS-base based on threshold according to the topology of the graph and the similarity between edges containing time tags.Furthermore,an optimization algorithm DTS-opt based on the fast calculation of maximum similarity time slice is proposed in order to accelerate the convergence speed.Finally,experiments on real data sets demonstrate the efficiency and scalability of the proposed algorithms.
KSN:A Web Service Discovery Method Based on Knowledge Graph and Similarity Network
YU Yang, XING Bin, ZENG Jun, WEN Jun-hao
Computer Science. 2021, 48 (10): 160-166.  doi:10.11896/jsjkx.200900026
Abstract PDF(2829KB) ( 758 )   
References | Related Articles | Metrics
Service discovery aims to solve the problem of service information explosion,find and locate services that meet the needs of service requesters.Since the service description information is mainly composed of short text with noise and has the feature of sparse semantics,it is difficult to extract the implicit context information of the service description document.In addition,the traditional service discovery method directly obtains the characteristic representation of the service.According to the cosine similarity to calculate the similarity,the used measurement function is not in line with human perception.In response to the above two problems,this paper proposes a service discovery framework (KSN) based on knowledge graphs and neural similar networks.It uses the knowledge graph to connect the entities in the service description and specifications to obtain rich external information,thereby enhancing the semantic information of the service description.And it uses convolutional neural network (CNN) to extract the feature vector of the service as the input of the neural similarity network.The neural similarity network will learn a similarity function to calculate the similarity between the service and the request to support the service discovery process.A large number of experiments on real service data sets crawled by ProgrammableWeb show that KSN is superior toexisting Web service discovery methods in terms of multiple evaluation metrics.
Temporal RDF Model and Index Method Based on Neighborhood Structure
CHEN Yuan-yuan, YAN Li, ZHANG Zhe-qing, MA Zong-min
Computer Science. 2021, 48 (10): 167-176.  doi:10.11896/jsjkx.200900114
Abstract PDF(2454KB) ( 609 )   
References | Related Articles | Metrics
Resource description framework (RDF) is a metadata model and information description specification recommended by W3C,which is widely used in various fields.To track changes in RDF data over time,temporal information is introduced into the RDF framework.With the rapid growth of temporal RDF data,effective management of temporal RDF data is necessary.A reasonable index mechanism can achieve efficient storage and query of data.In this paper,we first present a temporal RDF data mo-del.We propose a specific one-dimensional coding scheme,which represent temporal data simply and extend the existing RDF data model with lower overhead.Furthermore,we present its two levels of indexes based on neighborhood structure.The first one uses dynamic counting filter to index the neighborhood information of the node,and the second builds the B+ tree to index the temporal RDF data related to each node.Moreover,large-scale temporal RDF data can be updated.Experimental results show that the proposed method is around 35% better than the comparison method in most cases,and it is scalable and effective.
Urban Traffic Flow Completion with Multi-view Attention Mechanism
KANG Yan, CHEN Tie, LI Hao, YANG Bing, ZHANG Ya-chuan, BU Rong-jing
Computer Science. 2021, 48 (10): 177-184.  doi:10.11896/jsjkx.200800077
Abstract PDF(4018KB) ( 632 )   
References | Related Articles | Metrics
Traffic flow information is an important basis for intelligent transportation systems and urban computing.Traffic flow data is a new type of time series data.Due to the data collection method and the influence of external complex factors,the phenomenon of data loss is common and unavoidable.How to effectively mine the spatial-temporal characteristics of traffic flow data and the correlation between the data becomes the key to improve the missing data completion accuracy.Traditional statistical methods cannot meet the increasingly complex data requirements,and the application of deep learning promotes the development of missing data completion methods to higher accuracy.The article deeply analyzes the spatial-temporal characteristics of traffic flow,makes assumptions about the missing traffic flow,and proposes a UMAtNet (U-net with Multi-view Attention Mechanisms) traffic flow complement model.The model fuses closeness,trend and period time data with spatial data,and adopts diffe-rent data correlation measurement methods to fuse a multi-view attention mechanism,which can optimize the impact of the model on the spatial correlation of missing data.In order to verify the model,we use the open source data set of Beijing traffic data in the experiment,and analyzes in detail the influence of each part of the model and the loss function on the completion accuracy.The experimental results show that the fusion of UMAtNet and corresponding components can further improve the completion accuracy.
Sample Feature Kernel Matrix-based Sparse Bilinear Regression
SHAO Zheng-yi, CHEN Xiu-hong
Computer Science. 2021, 48 (10): 185-190.  doi:10.11896/jsjkx.200800219
Abstract PDF(1525KB) ( 549 )   
References | Related Articles | Metrics
There are a large number of redundant data in many real applications,which may be high dimensional.In this case,there will be many problems in regression prediction,such as overfitting and low prediction accuracy.In addition,most regression methods are based on vectors,ignoring the relationship between the original positions of matrix data.To this end,a sample kernel matrix-based sparse bilinear regression (KMSBR) method is proposed.The KMSBR model which use the sample feature kernel matrix and L2,1-norm is established through the left and right regression coefficient matrix.Thus,the KMSBR can implement the selection of samples and its features simultaneously.Experimental results on several data sets show that KMSBR can effectively select samples and its features,thus improve the efficiency of the algorithm,and the prediction accuracy is better than the existing regression models.
Gaussian Mixture Models Algorithm Based on Density Peaks Clustering
WANG Wei-dong, XU Jin-hui, ZHANG Zhi-feng, YANG Xi-bei
Computer Science. 2021, 48 (10): 191-196.  doi:10.11896/jsjkx.200800191
Abstract PDF(1846KB) ( 662 )   
References | Related Articles | Metrics
Due to the existence of a large number of sample data which obey the Gaussian distribution,GMM (Gaussian mixture models) is used to cluster these sample data and get more accurate clustering results.In general,EM algorithm(expectation maxi-mization algorithm) is used to estimate the parameters of GMM iteratively.However,the traditional EM algorithm has two shortcomings:it is sensitive to the initial clustering center;the itera-tive termination condition of iterative parameter estimation is to judge that the distance between two adjacent estimated parameters is less than a given threshold,which can't guarantee that the algorithm converges to the optimal value of the parameters.In order to overcome the above shortcomings,density peaks clustering (DPC) is proposed to initialize EM algorithm to improve the robustness of the algorithm.The relative entropy is used as the ite-ration termination condition of the EM algorithm to optimize the parameters of GMM algorithm.The comparative experiments on artificial datasets and UCI datasets show that the new algorithm not only improves the robustness of EM algorithm,but also outperforms the traditional clustering algorithm.On the datasets which obey Gaussian distribution,the new algorithm greatly improves the clustering accuracy.
Fuzzy Clustering Validity Index Combined with Multi-objective Optimization Algorithm and Its Application
CUI Guo-nan, WANG Li-song, KANG Jie-xiang, GAO Zhong-jie, WANG Hui, YIN Wei
Computer Science. 2021, 48 (10): 197-203.  doi:10.11896/jsjkx.200900061
Abstract PDF(2399KB) ( 701 )   
References | Related Articles | Metrics
Fuzzy clustering method can analyze complex data sets more effectively.Because there are many kinds of fuzzy clustering algorithms and the clustering results will change with the number of input clusters,the results of fuzzy clustering algorithm are not accurate,so the number of fuzzy clustering k must be determined in order to obtain certain clustering results.At present,the existing research mainly uses a variety of fuzzy clustering effectiveness indexes to determine the optimal number of clusters k.However,fuzzy clustering indexes such as SSD,PBM will decrease monotonically with the increase of clustering number k,which makes it impossible to determine the optimal number of clusters k.Therefore,this paper proposes a fuzzy clustering validity index (OSACF) combined with a multi-objective optimization algorithm,which combines fuzzy clustering validity with a multi-objective optimization algorithm (MOEA) to solve the optimal number of clusters k problem.Different from using clustering validity index,OSACF establishes a bi-objective model between cluster number k and clustering validity index,and uses MOEA to optimize the bi-objective model to determine the optimal cluster number k,so as to avoid the influence of monotonous decreasing of clustering validity index.On the other hand,OSACF uses morphological similarity distance to replace the traditional Euclidean distance metric,which avoids the influence of cluster shape on the calculation of cluster k.The experimental results show that the optimal fuzzy cluster number k obtained by OSACF combined with MOEA is more accurate than the results obtained by the existing clustering effectiveness indicators.
Computer Graphics & Multimedia
Multi-orientation Partitioned Network for Person Re-identification
TANG Yi-xing, LIU Xue-liang, HU She-jiao
Computer Science. 2021, 48 (10): 204-211.  doi:10.11896/jsjkx.210300128
Abstract PDF(2692KB) ( 492 )   
References | Related Articles | Metrics
Combining global features with local features is an important solution to improve discriminative performances in person re-identification (Re-ID) task.In the past,external information was used to locate regions with corresponding semantics,thus mining local information.Most of these methods are not end-to-end,so the training process is complex.To solve this problem,a multi-orientation partitioned network (MOPN) is proposed,which can effectively mine local information and combine global information with local information for end-to-end feature learning.The network has three branches:one for extracting global feature and two for mining local information.Without relying on external information,the algorithm divides pedestrians' images into hori-zontal and vertical stripes in different local branches respectively,so as to obtain different local feature representations.Plenty of experiments conducted on Market-1501,DukeMTMC-reID,CUHK03 and cross-modal dataset SketchRe-ID show that the proposed method has better overall performance than other comparison algorithms,and is effective and robust.
Light Field Depth Estimation Method Based on Encoder-decoder Architecture
YAN Xu, MA Shuai, ZENG Feng-jiao, GUO Zheng-hua, WU Jun-long, YANG Ping, XU Bing
Computer Science. 2021, 48 (10): 212-219.  doi:10.11896/jsjkx.200900005
Abstract PDF(4004KB) ( 501 )   
References | Related Articles | Metrics
Aiming at the solution to the time-consuming and low-precision disadvantage of present methodologies,the light field depth estimation method combining context information of the scene is proposed.This method is based on an end-to-end convolutional neural network,with the advantage of obtaining depth map from a single light field image.On merit of the reduced computational cost from this method,the time consumption is consequently decreased.For improvement in calculation accuracy,multi orientation epipolar plane image volumes of the light field images are input to network,from which feature can be extracted by the multi-stream encoding module,and then aggregated by the encoding-decoding architecture with skip connection,resulting in fuse the context information of the neighborhood of the target pixel in the process of per-pixel disparity estimation.Furthermore,the model uses convolutional blocks of different depths to extract the structural features of the scene from the central viewpoint image,by introducing these structural features into the corresponding skip connection,additional references for edge features are obtained and the calculation accuracy is further improved.Experiments in the HCI 4D Light Field Benchmark show that the BadPix index and MSE index of the proposed method are respectively 31.2% and 54.6% lower than those of the comparison me-thod,and the average calculation time of depth estimation is 1.2 seconds,which is much faster than comparison method.
Material Recognition Method Based on Attention Mechanism and Deep Convolutional Neural Network
XU Hua-jie, YANG Yang, LI Gui-lan
Computer Science. 2021, 48 (10): 220-225.  doi:10.11896/jsjkx.200800073
Abstract PDF(2071KB) ( 1005 )   
References | Related Articles | Metrics
The purpose of material recognition is to identify the main objects and their material categories in natural material images.Aiming at the problem of low recognition accuracy caused by the lack of data in material image data sets and the difficulty of manually labeling local texture regions,a material recognition method based on attention mechanism and deep convolutional neural network is proposed.The core of the method is material recognition deep convolutional neural network (MaterialNet).MaterialNet uses the deep residual network to extract the features of the image,and introduces the attention mechanism by the proposed cascaded atrous spatial pyramid pooling method,so that the network can adaptively focus on the key areas containing texture features through end-to-end training,so as to effectively identify the local texture features of materials.Based on the FMD material datasets,the experimental results show that the overall identification accuracy of MaterialNet is 82.3%,which is 7.2% and 4.5% higher than the current mainstream B-CNN and CNN+FV material identification methods,respectively.The recognition accuracy of MaterialNet is high for a variety of materials,and it has the advantages of less parameters and less calculation.
Camouflaged Object Detection Based on Improved YOLO v5 Algorithm
WANG Yang, CAO Tie-yong, YANG Ji-bin, ZHENG Yun-fei, FANG Zheng, DENG Xiao-tong, WU Jing-wei, LIN Jia
Computer Science. 2021, 48 (10): 226-232.  doi:10.11896/jsjkx.210100058
Abstract PDF(4547KB) ( 1796 )   
References | Related Articles | Metrics
Since the camouflage object is highly similar to the surrounding environment with a rather small size,the general detection algorithm is not fully applicable to the camouflaged object detection task,which makes the detection of camouflaged object more challenging than the general detection task.In order to solve this problem,the existing methods are analyzed in this paper and a detection algorithm for camouflage object is proposed based on the YOLO v5 algorithm.A new feature extraction network combined with attention mechanism is designed to highlight the feature information of the camouflage target.The original path aggregation network is improved so that the high,middle and lowly level feature map information is fully fused.The semantic information of the target is strengthened by nonlinear pool module,and the detection feature map size is increased to improve the detection recall rate of the small size target.On a public camouflage target dataset,the proposed algorithm is tested with 7 algorithms.The mAP of the proposed algorithm is 4.4% higher than that of the original algorithm,while the recall rate has improved 2.8%,which verifies the effectiveness of the algorithm for camouflaged object detection and the great advantage in accuracy compared with other algorithms.
Small Object Detection Oriented Improved-RetinaNet Model and Its Application
LUO Yue-tong, JIANG Pei-feng, DUAN Chang, ZHOU Bo
Computer Science. 2021, 48 (10): 233-238.  doi:10.11896/jsjkx.200900172
Abstract PDF(3292KB) ( 1054 )   
References | Related Articles | Metrics
Object detection algorithms based on deep learning are widely used in industrial detection.The RetinaNet algorithm has attracted much attention because of its advantages in both speed and accuracy.However,for small objects smaller than 32×32 pixels,the detection accuracy of this algorithm cannot meet the requirements of industrial detection.To this end,this article takes the enhancement of small object training as the basic idea,and makes the following improvements to the RetinaNet algorithm:in the sampling phase,the low-level feature map P2 is added to the FPN to ensure that the small object can be fully sampled,and adaptive training sample selection(ATSS) strategy is introduced to ensure that the detection speed is still fast enough after the feature layer is increased;the loss weight adjustment strategy is adopted in the later training stage to improve the fit of difficult samples in small objects.For the public data set MS COCO 2017 and the LED dispensing industrial data set in practical applications,the detection accuracy of this method for objects smaller than 32×32 increases by 4.1% and 10.7%,respectively,indicating that this method can significantly improve the detection ability of small objects.
Coherent Semantic Spatial-Temporal Attention Network for Video Inpainting
LIU Lang, LI Liang, DAN Yuan-hong
Computer Science. 2021, 48 (10): 239-245.  doi:10.11896/jsjkx.200600130
Abstract PDF(4565KB) ( 1046 )   
References | Related Articles | Metrics
Existing video inpainting methods usually produce blurred texture,distorted structure and artifacts,while applying image-based inpainting model directly to the video inpainting will lead to inconsistent time.From the perspective of time,a novel coherent semantic spatial-temporal attention(CSSTA) for video inpainting is proposed,through the attention layer,the model focuses on the information that the target frame is partially blocked and the adjacent frames are visible,so as to obtain the visible content to fill the hole region of the target frame.The CSSTA layer can not only model the semantic correlation between hole features but also remotely correlate the long-range information with the hole regions.In order to complete semantically coherent hole regions,a novel loss function Feature Loss is proposed to replace VGG Loss.The model is built on a two-stage coarse-to-fine encoder-decoder model for collecting and refining information from adjacent frames.Experimental results on the YouTube-VOS and DAVIS datasets show that the method in this paper runs almost in real-time and outperforms the three typical video inpainting methods in terms of inpainting results,peak signal-to-noise ratio (PSNR) and structural similarity (SSIM).
Information Security
Survey of Human Identification Algorithms Based on Wi-Fi Signal
KONG Jin-sheng, LI Jing-xin, DUAN Peng-song, CAO Yang-jie
Computer Science. 2021, 48 (10): 246-257.  doi:10.11896/jsjkx.210100076
Abstract PDF(2405KB) ( 906 )   
References | Related Articles | Metrics
In recent years,Wi-Fi perception has become an emerging research direction in human-computer interaction due to its advantages of low cost,non-contact,unaffected by light,and better privacy.At present,the research of Wi-Fi perception has expanded from target positioning to fields such as action recognition and identity recognition.Taking human identity recognition as an example,the research on Wi-Fi perception technology in this field is summarized and analyzed.First,it briefly summarizes the development history,advantages and disadvantages of Wi-Fi perception technology,and introduces the advantages of using Wi-Fi signals for identification compared with traditional identification technologies.Second,it introduces Wi-Fi in detail.The basic process of perceiving identity includes 4 steps of signal acquisition,preprocessing,feature extraction,and identity recognition,and the specific operation process of each step is introduced in detail.Third,the existing gait and gestures are compared to the existing ones in terms of gait and gestures.The research results of Wi-Fi-aware identity are compared and analyzed horizontally and vertically.Finally,in view of the key issues in the current research in this field,the focus of future research is proposed,mainly including multi-person identification and migration learning.
Feature Transformation for Defending Adversarial Attack on Image Retrieval
XU Xing, SUN Jia-liang, WANG Zheng, YANG Yang
Computer Science. 2021, 48 (10): 258-265.  doi:10.11896/jsjkx.200800222
Abstract PDF(3207KB) ( 694 )   
References | Related Articles | Metrics
The adversarial attack is firstly studied in image classification to generate imperceptible perturbations that can mislead the prediction of a convolutional neural network.Recently,it has also been extensively explored in image retrieval and shows that the popular image retrieval models are undoubtedly vulnerable to return irrelevant images to the query image with small perturbations.In particular,landmark image retrieval is a research hotspot of image retrieval as an explosive volume of landmark images are uploaded on the Internet by people using various smart devices when taking tours in cities.This paper makes the first trail to investigate the defending approach against adversarial attacks on city landmark image retrieval models without training.Specifica-lly,we propose to perform image feature transformation at inference time to eliminate the adversarial effects based on the basic image features.Our method explores four feature transformation schemes:resize,padding,total variance minimization and image quilting,which are performed on a query image before feeding it to a retrieval model.Our defense method has the following advantages:1) no fine-tuning and incremental training procedure is required,2) very few additional computations and 3) flexible ensembles of multiple schemes.Extensive experiments show that the proposed transformation strategies are advanced at defending the existing adversarial attacks performed on the state-of-the-art city landmark image retrieval models.
Location Privacy Game Mechanism Based on Generative Adversarial Networks
WEI Li-qi, ZHAO Zhi-hong, BAI Guang-wei, SHEN Hang
Computer Science. 2021, 48 (10): 266-271.  doi:10.11896/jsjkx.200900021
Abstract PDF(1778KB) ( 651 )   
References | Related Articles | Metrics
This paper proposes a user-centered location privacy game mechanism,which is aimed to generate corresponding protection strategies based on the LBS service quality,and reduce the calculation scale and utility loss.This mechanism is based on the Stackelberg game model.When a user requests a LBS service,he/she uses the location ambiguity mechanism to disturb his/her location and send it to the LBS server,making it difficult for the attacker to predict his/her real location.Based on part of their known background knowledge,attackers infer the protection policies of users in the anonymous area and adjust their attack methods to minimize the level of user privacy.In order to solve the problem of large scale and long time calculation by traditional mathematical methods,this paper adopts generating countermeasures network to participate in the generation of protection strategy,and reduces the utility cost as much as possible.The experimental results show that the protection mechanism has good performance in terms of privacy protection level,and at the same time,it significantly reduces the generation time of the protection mechanism while losing some quality of service.
Security Analysis and Improvement of Strongly Secure Certificateless Digital Signature Scheme
YE Sheng-nan, CHEN Jian-hua
Computer Science. 2021, 48 (10): 272-277.  doi:10.11896/jsjkx.201200117
Abstract PDF(1360KB) ( 704 )   
References | Related Articles | Metrics
Certificateless public key cryptosystem combines the advantages of identity-based cryptosystem and traditional PKI public key cryptosystem,overcomes the key escrow problem of identity-based public key cryptosystem and the certificate management problem of PKI system,and has obvious advantages.By analysing the security of a strongly secure certificateless signature scheme proposed by Hassouna,et al,it shows that the scheme cannot resist the attack of falsifying messages and do not use private key generated by system master key to sign.So it is not a certificateless signature scheme.On this basis,an improved certificateless signature scheme is proposed and it proves the scheme can resist the attack of the first class of strong adversaries and the second class of adversaries.In the random oracle model and under the assumption of the Diffie-Hellman problem of the elliptic curve,the improved scheme satisfies the existential forgery.
TopoObfu:A Network Topology Obfuscation Mechanism to Defense Network Reconnaissance
LIU Ya-qun, XING Chang-you, GAO Ya-zhuo, ZHANG Guo-min
Computer Science. 2021, 48 (10): 278-285.  doi:10.11896/jsjkx.210400296
Abstract PDF(2709KB) ( 617 )   
References | Related Articles | Metrics
Some typical network attacks,such as link-flooding attack,need to be carried out on critical links based on topology reconnaissance,which has strong destructiveness and stealthiness.In order to defense these attacks effectively,TopoObfu,a topology obfuscation mechanism against network reconnaissance,is proposed.TopoObfu can add virtual links to the real network according to the requirements of network topology obfuscation,and provide attacker with fake topology by modifying the forwar-ding rules of probing packets,and hide critical links in the network.To facilitate the implementation,TopoObfu maps the fake topology to the flow table entries used by SDN switches for packet processing,and can be deployed in the hybrid network where only part of the nodes are SDN switches.The simulation analysis based on several typical real network topologies shows that TopoObfu can effectively improve the difficulty of critical links analysis launched by attackers in terms of link importance,network structure entropy,path similarity and so on,and has high implementation efficiency in terms of the number of flow table entries in SDN switches,the generated time of fake topology,and can reduce the probability of critical links being attacked.
Neural Network-based Binary Function Similarity Detection
FANG Lei, WEI Qiang, WU Ze-hui, DU Jiang, ZHANG Xing-ming
Computer Science. 2021, 48 (10): 286-293.  doi:10.11896/jsjkx.200900185
Abstract PDF(1802KB) ( 1009 )   
References | Related Articles | Metrics
Binary code similarity detection has extensive and important applications in program traceability and security audit.In recent years,the application of neural network technology in binary code similarity detection has broken through the performance bottleneck encountered by traditional detection techno-logy in large-scale detection tasks,making code similarity detection technology based on neural network embedding gradually become a research hotspot.This paper proposes a neural network-based binary function similarity detection technology.This paper first uses a uniform intermediate representation to eliminate the diffe-rences in instruction architecture of assembly code.Secondly,at the basic block level,it uses a word embedding model in natural language processing to learn the intermediate representation code and obtain the basic block semantic embedding.Then,at the function level,it uses an improved graph neural network model to learn the control flow information of the function,taking consideration of the basic block semantics at the same time,and to obtain the final function embedding.Finally,the similarity between two functions is measured by calculating the cosine distance between the two function embeddingvectors.This paper also implements a prototype system based on this technology.Experiments show that the program code representation learning process of this technology can avoid the introduction of human bias,the improved graph neural network is more suitable for learning the control flow information of functions,and the scalability and detection accuracy of our system are both improved,compared with the existing schemes.
Multi-stage Game Based Dynamic Deployment Mechanism of Virtualized Honeypots
GAO Ya-zhuo, LIU Ya-qun, ZHANG Guo-min, XING Chang-you, WANG Xiu-lei
Computer Science. 2021, 48 (10): 294-300.  doi:10.11896/jsjkx.210500071
Abstract PDF(2547KB) ( 753 )   
References | Related Articles | Metrics
As an important deception defense method,honeypot is of great significance to enhance the network active defense capability.However,most of the existing honeypots adopt the static deployment method,which is difficult to deal with the strategic attacks effectively.Therefore,by combining the complete information static game with Markov decision process,we propose a multi-stage stochastic game based dynamic deployment mechanism HoneyVDep.By taking the resource constrained maximum comprehensive gain of the defensive side as the goal,HoneyVDep establishes a multi-stage random game based honeypot deployment optimization model.Besides,we also implement a Q_Learning based solution algorithm,which can deal with the attacker's strategic detection attack behavior quickly.Finally,based on software defined network and virtualization containers,we implement an extensible prototype system.The experimental results show that HoneyVDep can effectively generate honeypot deployment strategy according to the characteristics of the attacker's attack behavior,improve the trapping rate of the attackers,and reduce the deployment cost.
Private Set Intersection Protocols Among Multi-party with Cloud Server Aided
WANG Qin, WEI Li-fei, LIU Ji-hai, ZHANG Lei
Computer Science. 2021, 48 (10): 301-307.  doi:10.11896/jsjkx.210300308
Abstract PDF(1568KB) ( 1094 )   
References | Related Articles | Metrics
Private set intersection (PSI) is a secure multi-party computation technique that allows several parties,who each hold a set of private items,to compute the intersection of those private sets without revealing additional information.PSI has been widely used in the field of artificial intelligence security and data mining security.With the advent of the multi-source data sharing era,most PSI protocols mainly solve the problem of two-party privacy set intersection,which can not be directly extended to multi-party privacy intersection computing scenarios.This paper designs a multi-party privacy intersection protocol with the help of cloud servers,which can outsource a part of the computation and communication to untrusted cloud server without disclosing any privacy data.This paper makes the protocol more efficient by using the methods of oblivious pseudo-random functions,secret sharing and key-value pair packing.It proves that the PSI protocol can be secure in the semi-honest model and all participants and cloud servers can not obtain the additional data.Compared with the existing scheme,the proposed protocol has the merit of less restricted and more applicable in application scenarios.
Method of Malware Family Classification Based on Attention-DenseNet-BC Model Mechanism
LI Yi-meng, LI Cheng-hai, SONG Ya-fei, WANG Jian
Computer Science. 2021, 48 (10): 308-314.  doi:10.11896/jsjkx.210200166
Abstract PDF(4331KB) ( 600 )   
References | Related Articles | Metrics
Malware is one of the most serious threats to the Internet.The existing malware has huge data size and various features.Convolutional Neural Network has the features of autonomous learning,which can be used to solve the problems that the feature extraction of malware is complex and the feature selection is difficult.However,in convolutional neural network,conti-nuously increasing the network layers will cause a disappear of the gradient,leading to a degradation of network performance and low accuracy.To solve this problem,an Attention-DenseNet-BC model that is suitable for malware image detection is proposed.First,the Attention-DenseNet-BC model is constructed by combining the DenseNet-BC network and the attention mechanism.Then,the malware images are used as the input of the model,and the detection results are obtained by training and testing the model.The experimental results indicate that compared with other deep learning models,the Attention-DenseNet-BC model can achieve better classification results.A high classification accuracy can be attained with the model based on the malimg public dataset.
Interdiscipline & Frontier
Survey on Improvement and Application of Grover Algorithm
LIU Xiao-nan, SONG Hui-chao, WANG Hong, JIANG Duo, AN Jia-le
Computer Science. 2021, 48 (10): 315-323.  doi:10.11896/jsjkx.201100141
Abstract PDF(1491KB) ( 1225 )   
References | Related Articles | Metrics
Quantum information science is a new interdisciplinary subject,which has unique performance in the field of information.It can break through the limits of the existing classical information systems in the aspects of improving computing speed,ensuring information security,increasing information capacity and improving detection accuracy.Grover algorithm is a typical quantum algorithm,which can realize quadratic acceleration for any classical brute force exhaustive search problem,further promoting the development of quantum computing.How to effectively improve and apply Grover algorithm has become an important research field of quantum computing.This paper summarizes the optimization,improvement and application of Grover algorithm,summarizes the application and improvement of Grover algorithm in different fields,and discusses some research directions of future algorithm improvement and related applications of Grover algorithm.
Mechanism and Path of Optimizing Institution of Legislative Evaluation by Applying “Big Data+Blockchain”
ZHANG Guang-jun, ZHANG Xiang
Computer Science. 2021, 48 (10): 324-333.  doi:10.11896/jsjkx.201200105
Abstract PDF(2003KB) ( 469 )   
References | Related Articles | Metrics
Although the application of big data in legislative evaluation helps to get rid of chronic diseases such as single evaluation subject,data omission,data distortion and formalization of conclusions.However,there are still some defects,such as lack of public participation under the dominant mode of the legislature,improper cleaning methods leading to poor data quality,information asymmetry,trust crisis caused by “algorithm black box”,and biased evaluation conclusions caused by algorithm bias.Based on the approach of technology-system collaborative evolution theory,this paper applies the blockchain which is consistent with the nature of big data to innovate the underlying architecture,makes up for the defects in the application of big data in legislative evaluation,and then constructs an anastomosing application model of “big data+block chain”,optimizes the legislative evaluation system with public participation incentive mechanism,data cleaning and review mechanism,consensus formation negotiation platform and procedure justification circulation platform,so as to fully release the system function of legislative evaluation to gather public opinion and wisdom,strengthen democratic supervision and improve the quality of legislation.
Failure-resilient DAG Task Rescheduling in Edge Computing
CAI Ling-feng, WEI Xiang-lin, XING Chang-you, ZOU Xia, ZHANG Guo-min
Computer Science. 2021, 48 (10): 334-342.  doi:10.11896/jsjkx.210300304
Abstract PDF(3486KB) ( 704 )   
References | Related Articles | Metrics
By deploying computation and storage resources at the network edge that is close to the data source,and scheduling tasks offloaded by users efficiently,edge computing can greatly improve the quality of experience (QoE) of users.However,due to the lack of the reliable infrastructure support,the failure of edge servers or communication links could easily fail the edge computing service.To handle this problem,we establish the failure models of the computing nodes and communication links in edge computing,and then propose the rescheduling algorithm DaGTR (Dependency-aware Greedy Task Rescheduling) for the scheduling of dependent user tasks in resource failure scenarios.DaGTR includes two sub-algorithms,DaGTR-N and DaGTR-L,which are responsible for handling the node and link failure events respectively.DaGTR can sense the data dependency of tasks,and reschedule the tasks affected by failure events based on greedy method to ensure the successful execution of each task.Simulation results show that the algorithm can effectively avoid the task failure caused by failure events and improve the success rate of tasks in the case of resource failure.
Microservices User Requests Allocation Strategy Based on Improved Multi-objective Evolutionary Algorithms
ZHU Han-qing, MA Wu-bin, ZHOU Hao-hao, WU Ya-hui, HUANG Hong-bin
Computer Science. 2021, 48 (10): 343-350.  doi:10.11896/jsjkx.201100009
Abstract PDF(2643KB) ( 840 )   
References | Related Articles | Metrics
How to allocate concurrent user requests to a system based on a microservices architecture to optimize objectives such as time,cost,and load balance,is one of the important issues that microservices-based application systems need to pay attention to.The existing user requests allocation strategy based on fixed rules only focuses on the solving of load balance,and it is difficult to deal with the balance between multi-objective requirements.A microservices user requests allocation model with multiple objectives of total requests processing time,load balancing rate,and total communication transmission distance is proposed to study the allocation of user requests among multiple microservices instances deployed in different resource centers.The multi-objective evolutionary algorithms with improved initial solutions generation strategy,crossover operator and mutation operator are used to solve this problem.Through many experiments on data sets of different scales,it is shown that the proposed method can better handle the balance between multiple objectives and has better solving performance,compared with the commonly used multi-objective evolutionary algorithms and traditional methods based on fixed rules.
Study on Co-evolution of Underload Failure and Overload Cascading Failure in Multi-layer Supply Chain Network
LI Shu, YANG Hua, SONG Bo
Computer Science. 2021, 48 (10): 351-358.  doi:10.11896/jsjkx.200900144
Abstract PDF(4447KB) ( 751 )   
References | Related Articles | Metrics
The supply chain network is closely related to our lives,and the cascading failure of the supply chain network has always been a hot research topic.This paper proposes a multi-layer supply chain network mixed failure model,which can better simulate the process of real supply chain network collapse and provide a reference for preventing supply chain network collapse.By establishing the supply chain network model of upper-level supplier network overload cascade failure and lower-level retailer network underload failure,the vulnerability of the supply chain network is studied when the upper and lower networks are attacked through different attack strategies.In the case of a certain initial attack ratio,the upper-layer supplier network is more robust than the lower-layer retailer network.Under the same attack ratio,the scale of the network crash when deliberately attacking network nodes is larger than that of random attacks.When the upper-layer supplier network node is initially attacked,the thresholdof network collapse is lower,which is more prone to collapse.This paper verifies the validity of the model and provides a new research model for preventing the collapse of the supply chain network.