乐胖代购免代理版

Survey of Recommender Systems Based on Graph Learning

CHENG Zhang-tao, ZHONG Ting, ZHANG Sheng-ming, ZHOU Fan

Computer Science. 2022, 49 (9): 1-13. doi:10.11896/jsjkx.210900072

Abstract

PDF(2404KB) ( 2741 )

References | Related Articles | Metrics

Collaborative filtering is a widely used technique in current recommendation systems.It leverages the similarity between different users or items to retrieve interactive information between users and items and recommends new items for target users.In recent years,graph learning has gradually become an emerging recommendation paradigm due to its excellent perfor-mance and scalability in graph representation learning.This paper systematically reviews the most recent research on recommendation field from the perspective of graph learning.First,we provide a taxonomy that groups the current recommendation scenarios into two categories according to the data type used,including recommendation systems based on interactive information that leverage user-item interaction data as the main data source and auxiliary information-enhanced recommendation systems that incorporate social information associated with users and items as well as the knowledge graph information.Then,we review the main approaches,fundamental algorithms and critical difficulties of current recommendation models from the perspectives of random walk,graph representation learning and graph neural networks.Finally,we summarize the main challenges of graph learning methods in the field of recommendation system and outline the possible future research directions.

Survey of Concept Drift Handling Methods in Data Streams

CHEN Zhi-qiang, HAN Meng, LI Mu-hang, WU Hong-xin, ZHANG Xi-long

Computer Science. 2022, 49 (9): 14-32. doi:10.11896/jsjkx.210700112

Abstract

PDF(2484KB) ( 1651 )

References | Related Articles | Metrics

At present,concept drift in the nonstationary data stream presents a trend of different speeds and and different space distribution,which has brought great challenges to many fields such as data mining and machine learning.In the past two de-cades,many methods dedicated to handling concept drift in nonstationary data streamsemerged.A novel perspective is proposed to classify these methods.The current concept drift handling methods are comprehensively explained from the explicit method of actively detection and the implicit method of passively adaption.In particular,active detection methods are analyzed from the per-spective of handling one specific type of concept drift and handling multiple types of concept drift,and passive adaptive methods are analyzed from the perspectives of single learner and ensemble learning.Many concept drift handling methods are analyzed and summarized in terms of the comparison algorithm,learning model,applicable drift type,advantages and disadvantages of algorithms.Finally,further research directions are given,including the concept drift handling methods in class-imbalanced data streams,the concept drift handling methods in data stream with the existence of novel classes,and the concept drift handling methods in the data stream with noise.

Generative Link Tree:A Counterfactual Explanation Generation Approach with High Data Fidelity

WANG Ming, WU Wen-fang, WANG Da-ling, FENG Shi, ZHANG Yi-fei

Computer Science. 2022, 49 (9): 33-40. doi:10.11896/jsjkx.220300158

Abstract

PDF(3156KB) ( 915 )

References | Related Articles | Metrics

The super large data scale and complex structure of deep models show excellent performance in processing and application of Internet data,but reduce the interpretability of AI systems.Counterfactual Explanations(CE) has received much attention from researchers as a special kind of explanation approach in the field of interpretability research.Counterfactual Explanations can be regarded as a kind of generated data in addition to being an explanation.From the viewpoint of application,this paper proposes an approach for generating counterfactual explanations with high data fidelity,called generative link tree(GLT),which uses a partitioning strategy and a local greedy strategy to construct counterfactual explanations based on the cases appearing in the training data.Moreover,it summarizes the generation methods of counterfactual explanations and select popular datasets to verify the GLT method.In addition,the metric of “Data Fidelity (DF)” is proposed to evaluate the fidelity and potential application of the counterfactual explanation as data from an experimental perspective.Compared with the baseline method,the data fidelity of the counterfactual explanation generated by the GLT method is significantly higher than that of the counterfactual explanation gene-rated by the baseline model.

Cross-domain Recommendation Based on Review Aspect-level User Preference Transfer

ZHANG Jia, DONG Shou-bin

Computer Science. 2022, 49 (9): 41-47. doi:10.11896/jsjkx.220200131

Abstract

PDF(2388KB) ( 945 )

References | Related Articles | Metrics

In order to solve the user cold-start problem caused by data-sparse in recommender system,this paper proposes a cross-domain recommendation algorithm based on aspect-level user preference transfer,named CAUT.CAUT is devised to learn aspect transfer across domains from a two-stage generative adversarial network and extract aspect-level user fine-grained prefe-rence from reviews.The data distribution misalignment between source and target domains is eliminated by fixing source domain encoder parameters and designing a domain discriminator.Then the user cold-start problem caused by data-sparse in the target domain could be alleviated by utilizing source domain data via CAUT.Experiments on real-world datasets show that the proposed CAUT outperforms SOTA models significantly in rating prediction RMSE indicator,suggesting that CAUT can effectively solve the user cold-start problem.

Collaborative Filtering Recommendation Method Based on Vector Quantization Coding

WANG Guan-yu, ZHONG Ting, FENG Yu, ZHOU Fan

Computer Science. 2022, 49 (9): 48-54. doi:10.11896/jsjkx.210700109

Abstract

PDF(2538KB) ( 613 )

References | Related Articles | Metrics

With the rapid development of the Internet,the emergence of massive data makes recommender system become a research hotspot in the field of computer science.Variational autoencoders(VAE) have been successfully applied to the design of collaborative filtering methods and achieved excellent recommendation results. However,there are some defects in the previous VAE-based models,such as the problems of prior constraint and the “posterior collapse”,which essentially reduce their recommendation performance.To address this issue while enabling the latent variable model more suitable for the recommendation task,a novel collaborative filtering recommendation model based on latent vector quantization is proposed in this paper.By encoding the discrete vectors instead of directly sampling from the distribution of latent variables,our method can learn discrete representations that are consistent with the observed data,which greatly improves the capability of latent vector encoding and the learning ability of the model.Extensive evaluations conducted on three benchmark datasets demonstrate the effectiveness of the proposed model.Our model can significantly improve the recommendation performance compared with existing state-of-the-art methods while learning more expressive latent representations.

Sequence Recommendation Based on Global Enhanced Graph Neural Network

ZHOU Fang-quan, CHENG Wei-qing

Computer Science. 2022, 49 (9): 55-63. doi:10.11896/jsjkx.210700085

Abstract

PDF(2660KB) ( 818 )

References | Related Articles | Metrics

Most of the existing session based recommendation systems recommend based on the correlation between the last clicked item and the user preference of the current session,and ignore that there may be item transitions related to the current session in other sessions,while these item transitions may also have a certain impact on users' current preferences Hence,it is indispensable to analyze users' preferences comprehensively from the perspective of local session and global session.Furthermore,most of these recommendation systems ignore the importance of location information,whereas items closer to the predicted location may be more relevant to the current user's interests.To solve these problems,this paper proposes a recommendation model based on global enhanced graph neural network with LSTM(GEL-GNN).GEL-GNN aims to predict the behavior of users according to all sessions,and GNN is employed to capture the global and local relationship of the current session,while LSTM is employed to capture the relationship between sessions at the global level.Firstly,users' preferences are to be translated as a combination of conversation interests based on global and local levels through the attention mechanism layer.Then,the distance between the current position and the predicted position is measured with the reverse position information,so that user behavior can be predicted more accurately.A number of experiments are conducted on three real data sets.Experimental results show that GEL-GNN is superior to the existing session-based graph neural network recommendation models.

Scientific Paper Heterogeneous Graph Node Representation Learning Method Based onUnsupervised Clustering Level

SONG Jie, LIANG Mei-yu, XUE Zhe, DU Jun-ping, KOU Fei-fei

Computer Science. 2022, 49 (9): 64-69. doi:10.11896/jsjkx.220500196

Abstract

PDF(2343KB) ( 752 )

References | Related Articles | Metrics

Knowledge representation of scientific paper data is a problem to be solved,and how to learn the representation of paper nodes in scientific paper heterogeneous network is the core to solve this problem.This paper proposes an unsupervised cluster-level scientific paper heterogeneous graph node representation learning method(UCHL),aiming at obtaining the representation of nodes (authors,institutions,papers,etc.) in the heterogeneous graph of scientific papers.Based on the heterogeneous graph representation,this paper performs link prediction on the entire heterogeneous graph and obtains the relationship between the edges of the nodes,that is,the relationship between paper and paper.Experiments results show that the proposed method achieves excellent performance on multiple evaluation metrics on real scientific paper datasets.

Aerial Target Grouping Method Based on Feature Similarity Clustering

CHAI Hui-min, ZHANG Yong, FANG Min

Computer Science. 2022, 49 (9): 70-75. doi:10.11896/jsjkx.210800203

Abstract

PDF(2315KB) ( 792 )

References | Related Articles | Metrics

In order to solve the problems that the number of clusters needs to be given and the sensitivity to the initial positions of the cluster centers while clustering algorithm is utilized for target grouping,a novel aerial target grouping method based on feature similarity clustering is proposed.Firstly,the similarity between targets is calculated and the similarity matrix is constructed.Then,the connected branches of the similarity matrix are calculated to obtained the group center structure and the isolated target points are detected.The number of group center structures is the number of clusters.Finally,the targets which are not belonging to the group center structure and the isolated points are clustered into the closest group center structure.It makes the clustering process no longer depend too much on the initialization of the cluster centers.Experimental results show that the proposed methodcan correctly identify the group center structure and detect the isolated points.Furthermore,its the clustering accuracy is higherthan that of other four clustering algorithms.

Author’s Academic Behavior Prediction Based on Heterogeneous Network Representation Learning

HUANG Li, ZHU Yan, LI Chun-ping

Computer Science. 2022, 49 (9): 76-82. doi:10.11896/jsjkx.210900078

Abstract

PDF(3031KB) ( 652 )

References | Related Articles | Metrics

The author's academic behavior prediction aims to mine the behavioral relationships of authors from heterogeneous academic networks to promote scientific research cooperation and produce high-level and high-quality research results.Most of the existing methods of node representation learning do not consider the semantic feature,content feature,global structure of the node,etc.It is difficult to effectively learn the low-dimensional characteristics of the node in the network.In order to effectively integrate the multi-dimensional features and global structure of nodes,a heterogeneous network representation learning method(HNEMA) that integrates BiLSTM,attention mechanism and clustering algorithm is proposed to improve the predictive effect of author's academic behavior.HNEMA first integrates the multi-dimensional features of nodes based on BiLSTM and attention mechanism,aggregates the same type of neighbors on the same meta-path or different meta-paths,and then aggregates the multi-dimensional features of all neighbors of the node to be characterized.Based on this,a clustering algorithm is used to capture the global features of the node,so as to comprehensively and effectively learn the low-dimensional characteristics of the node.On the basis of comprehensive feature learning,logistic regression classifier is used to predict author's academic behavior.Validation experiments on three public datasets show that HNEMA has a certain degree of improvement in AUC and F1 indicators compared to other methods.

Community Detection Algorithm Based on Node Stability and Neighbor Similarity

ZHENG Wen-ping, LIU Mei-lin, YANG Gui

Computer Science. 2022, 49 (9): 83-91. doi:10.11896/jsjkx.220400146

Abstract

PDF(3729KB) ( 780 )

References | Related Articles | Metrics

With the increase of the scale of complex network,community structure becomes more complex.The relationship between nodes and communities become more diversified.It is expected to improve community detection algorithm performance by effectively measuring the community structure and dealing with the nodes with different certainty of community belonging.This paper proposes a community detection algorithm based on node stability and neighbor similarity.Firstly,label entropy of nodes is defined to measure node stability and the nodes with low label entropy are selected as stable node sets.Then the neighbor simila-rity is defined according to the label of node neighbor and the community belonging consistency of nodes and their neighbors is measured.The initial network is constructed by using the node with the highest neighbor similarity between the stable node and its neighbor,and the initial community detection results with high reliability are obtained by running label propagation algorithm on the subnetwork.The unclustered nodes are allocated to the community of the node with the highest Katz similarity.The final result of community detection is obtained by merging small-scale communities.Compared with LPA,BGLL,Walktrap,Infomap,LPA-S and other classical algorithms,experimental results show that the NSNSA algorithm performs well in modularity and NMI.

Short Texts Feautre Enrichment Method Based on Heterogeneous Information Network

LYU Xiao-feng, ZHAO Shu-liang, GAO Heng-da, WU Yong-liang, ZHANG Bao-qi

Computer Science. 2022, 49 (9): 92-100. doi:10.11896/jsjkx.210700241

Abstract

PDF(2541KB) ( 552 )

References | Related Articles | Metrics

With the deep integration of computer technology into social life,more and more short text messages are spreaded all over the web platform.Aiming at the problem of data sparsity of short texts,a robust heterogeneous information network framework(HTE) for modeling short texts,which can integrate any type of additional information and capture the relationship between them to solve the data sparsity problem,is constructed.Based on this framework,six short text expansion methods are designed using different external knowledge,and the short text features are enriched by introducing entity information such as entities,entity categories,inter-entity relationships and textual information such as text topics from Wikipedia and Freebase knowledge bases.Finally,the similarity measurement result is used to verify the experimental effect.By comparing the six text expansion me-thods with the traditional three similarity measures on two short text datasets and the current mainstream short text matching algorithms,the results of the proposed six text expansion methods are improved.Compared with BERT,the similarity measurement results of the best method improves by 5.97%.The proposed framework is robust and can include any type of external know-ledge,and the proposed method can overcome the data sparsity problem of short texts and can perform similarity metrics on short texts with high accuracy in an unsupervised manner.

Time Series Data Anomaly Detection Based on Total Variation Ratio Separation Distance

XU Tian-hui, GUO Qiang, ZHANG Cai-ming

Computer Science. 2022, 49 (9): 101-110. doi:10.11896/jsjkx.210600174

Abstract

PDF(4093KB) ( 629 )

References | Related Articles | Metrics

Anomaly detection for time series data is one of the important research problems in data analysis.Its main challenge is to detect if there are any anomalies and locate anomalies with low delay according to context.Most of existing anomaly detection methods capture anomalies using the probability density ratio to measure similarity between sequences.These methods need to use the cross-validation method to estimate the parameters of probability density ratio.However,cross-validation can increase the computational complexity,resulting in low computational efficiency and a high time delay.To address these issues,this paper proposes a detection method based on total variation ratio separation distance,in which total variation is adopted to extract sequence fluctuation features.Due to the fact that the total variation ratio is better than probability density ratio,the proposed method achieves higher computational efficiency and lower time delay.To reduce noise interference and further improve the detection accuracy,the proposed method is combined with the relative total variation.Experimental results show that the proposed method performs well in terms of detection accuracy,low delay and computational efficiency.

Overview of Natural Language Video Localization

NIE Xiu-shan, PAN Jia-nan, TAN Zhi-fang, LIU Xin-fang, GUO Jie, YIN Yi-long

Computer Science. 2022, 49 (9): 111-122. doi:10.11896/jsjkx.220500130

Abstract

PDF(2218KB) ( 864 )

References | Related Articles | Metrics

Natural language video localization(NLVL),which aims to locate a target moment from a video that semantically corresponds to a text query,is a novel and challenging task.Different from the task of temporal action localization,NLVL is more flexible without restrictions from predefined action categories.Meanwhile,NLVL is more challenging since it requires align semantic information from both visual and textual modalities.Besides,how to obtain the final timestamp from the alignment relationship is also a tough task.This paper first proposes the pipeline of NLVL,and then categorizes them into supervised and weakly-supervised methods according to whether there is supervised information,following by the analysis of the strengths and weaknesses of each kind of method.Subsequently,the dataset,evaluation protocols and the general performance analysis are presented.Finally,the possible perspectives are obtained by summarizing the existing methods.

Fine-grained Semantic Reasoning Based Cross-media Dual-way Adversarial Hashing Learning Model

CAO Xiao-wen, LIANG Mei-yu, LU Kang-kang

Computer Science. 2022, 49 (9): 123-131. doi:10.11896/jsjkx.220600011

Abstract

PDF(3600KB) ( 711 )

References | Related Articles | Metrics

Cross-media hashing has received extensive attention in cross-media searching tasks due to its superior searching efficiency and low storage cost.However,existing methods cannot adequately preserve the high-level semantic relevance and multi-label of multi-media data.In order to solve the above problems,this paper proposes a fine-grained semantic reasoning based cross-media dual-way adversarial hashing learning model(SDAH),which generates compact and consistent cross-media unified efficient hash semantic representations by maximizing fine-grained semantic associations between different medias.First,a fine-grained cross-media semantic association learning and inference method based on the cross-media collaborative attention mechanism is proposed.The cross-media attention mechanism collaboratively learns the fine-grained implicit semantic associations of images and texts,and obtains the salient semantic inference features of images and texts.Then,a cross-media dual-way adversarial hashing network is established to jointly learn the intra-modality and inter-modality semantic similarity constraints,and better to align the semantic distributions of different media hash codes through a two-way adversarial learning mechanism,which generates higher-quality and more discriminative cross-media unified hash representation,facilitates the process of cross-media semantic fusion and improves the cross-media searching performance.Experimental results compared with existing methods on two public datasets verify the performance superiority of the proposed method in various cross-media search scenarios.

Dual Variational Multi-modal Attention Network for Incomplete Social Event Classification

ZHOU Xu, QIAN Sheng-sheng, LI Zhang-ming, FANG Quan, XU Chang-sheng

Computer Science. 2022, 49 (9): 132-138. doi:10.11896/jsjkx.220600022

Abstract

PDF(2303KB) ( 577 )

References | Related Articles | Metrics

The rapid development of the Internet and the continuous expansion of social media have brought a wealth of social event information,and the task of social event classification has become increasingly challenging.Making full use of image-level and text-level information is the key to social event classification.However,most of existing methods have the following limitations:1) Most of the existing multi-modal methods have an ideal assumption that the samples of each modality are sufficient and complete,but in real applications this assumption does not always hold and there will be cases where a certain modality of events is missing;2) Most methods simply concatenate image features and text features of social events to obtain multi-modal features to classify social events.To address these challenges,this paper proposes a dual variational multi-modal attention network(DVMAN) for social event classification to address the limitations of these existing methods.In the DVMAN network,this paper proposes a novel dual variational autoencoders network to generate public representations of social events and further reconstruct the missing modal information in incomplete social event learning.Through distribution alignment and cross-reconstruction alignment,image and text latent representations are doubly aligned to mitigate the gap between different modalities,and for the mis-sing modality information,a generative model is utilized to synthesize its latent representations.In addition,this paper designs a multi-modal fusion module to integrate the fine-grained information of images and texts of social events,so as to realize the complementation and enhancement of information between modalities.This paper conducts extensive experiments on two publicly available event datasets,compared with the existing advanced methods,the accuracy of DVMAN improves by more than 4%.It demonstrates the superior performance of the proposed method for social event classification.

Cross-image Text Reading Method Based on Text Line Matching

DAI Yu, XU Lin-feng

Computer Science. 2022, 49 (9): 139-145. doi:10.11896/jsjkx.220600032

Abstract

PDF(3493KB) ( 613 )

References | Related Articles | Metrics

Reading text with a camera can help the computer understand the text content.However,due to the limited field of view of the camera and the complexity of Chinese text recognition,it is sometimes difficult for the computer to read complete text content from a single text image with the camera.Thus,we define the cross-image text reading task,which aims to read the complete text content of a pair of overlapping text images.For the cross-image text reading task,we propose the cross-image text reading method via text line matching.We first adopt a text detection network to crop text lines.Then,we design the text line matching network with the multi-head self-attention mechanism to predict the matching relationships of text lines.Finally,the editing-based text reading network is proposed to remove overlapping texts and read complete text content.We also construct the cross-image Chinese text reading(CCTR) dataset for training and evaluation.Experiment results on CCTR dataset demonstrate that the proposed method achieves higher reading performance than the pixel-level stitching and recognition methods,which proves the superiority of the proposed method.

Study on Information Perception Based User Presence in Virtual Reality

QU Qian-wen, CHE Xiao-ping, QU Chen-xin, LI Jin-ru

Computer Science. 2022, 49 (9): 146-154. doi:10.11896/jsjkx.220500200

Abstract

PDF(3088KB) ( 866 )

References | Related Articles | Metrics

Virtual Reality(VR) technology builds a simulation environment through a computer,provides users with a three-dimensional dynamic view,and enhances the user's sensory experience,so that the user will get an immersive sense of immersion.With the rise and continuous development of virtual reality technology,people's visual and auditory experience has made great progress.With the development of multimedia technology,panoramic video emerges gradually.Compared with ordinary video,panoramic video has a wider viewing angle and richer visual information.The wide application of virtual reality technology and the development of panoramic video technology make virtual reality panoramic video(VR Video) become one of the most popular and concerned VR services.The information perception and acceptance behavior of users in the virtual reality environment have also been affected.Based on SMOTE algorithm,Bayesian network,logistic regression,and other statistical analysis methods and Few-shot Learning algorithm,this paper compares the differences in users' information memory degree,sense of reality,sense of participation,and other aspects when using VR head-mounted display and ordinary display iPad to watch panoramic video respectively.To explore the difference between the user's information acceptance effect and the sense of presence in the virtual reality environment and the traditional media environment.Experiments show that users with virtual reality headsets score an average of about 0.786 for message acceptance,compared with 0.634 for iPad users with regular displays.Among them,the score of positive information and non-positive information receiving effect in the virtual reality environment is 1.626 times and 1.245 times that of in the traditional environment respectively.However,it is noteworthy that in video A,which has the longest average scene duration,there is no significant difference in the information acceptance effect.In addition,the length of the user's visual residence time has a positive impact on the acceptance effect of information.After verifying that the sense of presence can be subdivided into the sense of reality and sense of participation,this paper shows that the score of users for the sense of presence in the iPad environment is always lower than that of in a VR environment through feature engineering and classification decision tree.After that,the correlation coefficient of the least square method is used to prove that presence has a positive effect on user information acceptance.Meanwhile,with the help of the Few-shot Learning algorithm,the average information memory number of video A－D is 9.20,9.13,8.83,and 10.57 when the sense of presence is strong,while the average information memory number of video A－D is 8.53,6.80,7.14 and 7.66 when the sense of presence is weak.It is proved that the user's sense of presence during information perception is beneficial to the user to obtain a better information acceptance effect.

Sequence-to-Sequence Chinese Continuous Sign Language Recognition and Translation with Multi- layer Attention Mechanism Fusion

ZHOU Le-yuan, ZHANG Jian-hua, YUAN Tian-tian, CHEN Sheng-yong

Computer Science. 2022, 49 (9): 155-161. doi:10.11896/jsjkx.210800026

Abstract

PDF(2632KB) ( 909 )

References | Related Articles | Metrics

Enabling computers to understand the expressions of signers has been a challenging task that requires considering not only the temporal and spatial information of sign language videos,but also the complexity of sign language grammar.In the continuous sign language recognition task,sign language words and sign language actions share a consistent order.In contrast,in the continuous sign language translation task,the generated natural language sentences have to conform to the spoken description,and the word order may not coincide with the action order.To enable more accurate learning of signers' expressions,this paper proposes a novel deep neural network for simultaneous sign language recognition and translation.In this scheme,we explore the effectiveness of different classical pre-trained convolutional neural networks,and different multilayer temporal attention score functions on continuous sign language recognition,combined with Transformer language model,to obtain continuous sign language translation conforming to the spoken description based on continuous sign language recognition.First,this method is assessed on the first large-scale complex background Chinese continuous sign language recognition and translation dataset Tslrt.The complex contextual environment and rich action expressions of signers in Tslrt dataset are used to train our neural network model through different comparison experiments,resulting in a series of benchmark results.The best WER are 4.8% and 5.1% on the tasks of continuous sign language recognition and translation,respectively.To further demonstrate the effectiveness of our method,experiments are conducted on another Chinese continuous sign language recognition dataset Chinese-CSL and compared with other 13 methods.The results show that the WER of our method reaches 1.8%,which proves the effectiveness of the proposed method.

Temporal Knowledge Graph Representation Learning

XU Yong-xin, ZHAO Jun-feng, WANG Ya-sha, XIE Bing, YANG Kai

Computer Science. 2022, 49 (9): 162-171. doi:10.11896/jsjkx.220500204

Abstract

PDF(1811KB) ( 2503 )

References | Related Articles | Metrics

As a structured form of human knowledge,knowledge graphs have played a great supportive role in supporting the semantic intercommunication of massive,multi-source,heterogeneous data,and effectively support tasks such as data analysis,attracting the attention of academia and industry.At present,most knowledge graphs are constructed based on non-real-time static data,without considering the temporal characteristics of entities and relationships.However,data in application scenarios such as social network communication,financial trade,and epidemic spreading network are highly dynamic and exhibit complex temporal properties.How to use time series data to build knowledge graphs and effectively model them is a challenging problem.Recently,numerous studies use temporal information in time series data to enrich the characteristics of knowledge graphs,endowing know-ledge graphs with dynamic features,expanding fact triples into quadruple representation(head entity,relationship,tail entity,time).The knowledge graph which utilizes time-related quadruples to represent knowledge are collectively referred to as temporal knowledge graph.This paper summarizes the research work of temporal knowledge graph representation learning by sorting out and analyzing the corresponding literature.Specifically,it first briefly introduce the background and definition of temporal know-ledge graph.Next,it summarizes the advantages of the temporal knowledge graph representation learning method compared with the traditional knowledge graph representation learning method.Then it elaborates on the recent method of temporal knowledge graph representation learning from the perspective of the method modeling facts,introduces the dataset used by the above method and summarizes the main challenges of this technology.Finally,the future research direction is prospected.

Overview of Multi-agent Deep Reinforcement Learning Based on Value Factorization

XIONG Li-qin, CAO Lei, LAI Jun, CHEN Xi-liang

Computer Science. 2022, 49 (9): 172-182. doi:10.11896/jsjkx.210800112

Abstract

PDF(2660KB) ( 1299 )

References | Related Articles | Metrics

Multi-agent deep reinforcement learning based on value factorization is one of many multi-agent deep reinforcement learning algorithms,and it is also a research hotspot in the field of multi-agent deep reinforcement learning.Under some constraints,the joint action value function of multi-agent system is factorized into a certain combination of individual action value function,which is able to effectively solve the problems of environment instability and exponential explosion of action space in multi-agent system.Firstly,this paper explains why value function factorization should be carried out and introduces the basic theory of multi-agent deep reinforcement learning.Secondly,according to whether to introduce other mechanisms and the diffe-rence of introduced mechanism,multi-agent deep reinforcement learning(MADRL)algorithm based on value factorization is divi-ded into three categories:simple factorization type,based on the individual-global-max(IGM)principle and based on attention mechanism.Then,according to the classifications,this paper emphatically introduces several typical algorithms and compares and analyzes their strengths and weaknesses.Finally,it briefly describes the application and development prospect of these algorithms.

Federated Learning Based on Stratified Sampling Optimization for Heterogeneous Clients

LU Chen-yang, DENG Su, MA Wu-bin, WU Ya-hui, ZHOU Hao-hao

Computer Science. 2022, 49 (9): 183-193. doi:10.11896/jsjkx.220500263

Abstract

PDF(4712KB) ( 732 )

References | Related Articles | Metrics

Federated learning(FL) is a new distributed learning framework for privacy protection,which is different from traditional distributed machine learning:1)differences in communication,computing,and storage performance among devices(device heterogeneity),2)differences in data distribution and data volume(data heterogeneity),and 3)high communication consumption.Under heterogeneous conditions,the data distribution of clients varies greatly,which leads to the decrease of model convergence speed.Especially in the case of highly heterogeneous condition,the traditional FL algorithm cannot converge and the training loss curve will fluctuate greatly with the increase of local iterations.In this work,a FL algorithm based on stratified sampling optimization(FedSSO) is proposed.In FedSSO,a density-based clustering method is used to divide the overall client into different clusters.Then,some available clients are proportionally extracted from different clusters to participate in training.Therefore,various data are involved in each training round to ensure that FL can accelerate convergence to the optimal solution.The strategy of learning rate decay and the choice of local iterations is set to ensure the convergence.The convergence of FedSSO algorithm is proved theoretically and experimentally,andthe superiority of FedSSO is demonstrated by comparing it with other FL algorithms on public MNIST,Cifar-10,and Sentiment140 datasets.

Collision Avoidance Planning for Unmanned Aerial Vehicles Based on Spatial Motion Constraints

LUO Xiong-feng, ZHAI Xiang-ping

Computer Science. 2022, 49 (9): 194-201. doi:10.11896/jsjkx.210700107

Abstract

PDF(2101KB) ( 673 )

References | Related Articles | Metrics

In three-dimensional space,how to conduct motion planning when unmanned aerial vehicles(UAV)facing moving obstacles is an interesting research direction.In dynamic environments,traditional algorithms based on the speed obstacles mainly aim at two-dimensional robots,which realizes obstacle avoidance actions by selecting velocities from the reachable velocity set out of the collision velocity set.This paper generalizes algorithms based on the current positions and velocities of UAV and obstacles to calculate the set of velocities that will cause collisions.According to the maximum speed and maximum acceleration of UAV,the velocity space that can be reached at the current moment is restricted.Different strategies are formulated for the needs of various scenes and are selected in a specific method in this subtraction set,so as to avoid obstacles in a three-dimensional scene with specific requirements.Aiming at the scene where UAV abstracted as spherical travels to the destination by avoiding spherical obstacles in 3D space,this paper verifies the obstacle avoidance algorithm by combining C++ and blueprint programming.It captures the movement trajectory of different strategies and records the corresponding consumption time,which demonstrates that the proposed algorithm can effectively complete the dynamic obstacle avoidance task of UAV in three-dimensional space.

Key-Value Relational Memory Networks for Question Answering over Knowledge Graph

RAO Zhi-shuang, JIA Zhen, ZHANG Fan, LI Tian-rui

Computer Science. 2022, 49 (9): 202-207. doi:10.11896/jsjkx.220300277

Abstract

PDF(2036KB) ( 810 )

References | Related Articles | Metrics

Question answering over knowledge graph(KG-QA) systems map the natural language question to the 〈subject,predicate,object〉 triple in the knowledge graph(KG) by semantic analysis of the given question,and infer the triple to get the answer of the question.Due to the diversity of natural languages,a question may be expressed in multiple forms but the triples in KGs are structured data in a standard form.It is challenging to map questions to triples in KGs.This paper proposes a novel Key-Value relational memory network,starting from the perspective of KGs,and focusing on the relationship between the candidate answer knowledge and the relationship between the knowledge in KGs and the question representations.In addition,the attention mechanism is applied in the proposed model,so that it has better interpretability than other baseline models.We evaluate the method on WebQuestions benchmark.Experiment results show that,compared with the best methods based on information extraction,the F1 value of the proposed method increases by 5.9% and is slightly higher than that of the optimal methods based on semantic analysis,which verifies the effectiveness of the proposed method.

Automated Container Terminal Oriented Travel Time Estimation of AGV

LENG Dian-dian, DU Peng, CHEN Jian-ting, XIANG Yang

Computer Science. 2022, 49 (9): 208-214. doi:10.11896/jsjkx.210700028

Abstract

PDF(2021KB) ( 675 )

References | Related Articles | Metrics

Automated guided vehicles(AGV)are crucial for the horizontal transportation of automated container terminals.Accurate estimation of the travel time of each AGV will reduce the number of idle AGV resources and increase the efficiency of the entire terminal.This paper proposes a method for travel time estimation of AGV in automated container terminals.Firstly,the target route of AGV is divided and encoded into several segments.Secondly,other routes are encoded as environment information,which depart before or after the target route.And the conflict between these routes and target route is estimated as an auxiliary task.Finally,the travel time with all encodings is calculated.The proposed method introduces the influence of path conflicts on time estimation.Experiments based on historical data of automated terminals show that,compared with static time estimation methods commonly used in AGV scenarios,the proposed method can reduce the time estimation error by more than 18%,and can estimate the travel time more accurately.

Ontology Alignment Method Based on Self-attention

WU Zi-yi, LI Shao-mei, JIANG Meng-han, ZHANG Jian-peng

Computer Science. 2022, 49 (9): 215-220. doi:10.11896/jsjkx.210700190

Abstract

PDF(2015KB) ( 595 )

References | Related Articles | Metrics

With the development of knowledge graph in the field of artificial intelligence,there is an increasing demand to integrate knowledge graph from different sources to obtain a big knowledge graph with wider coverage.Ontology is the superstructure that can guide the construction of knowledge graph.To solve the problem of ontology alignment in knowledge graph fusion,this paper proposes an ontology alignment method based on self-attention model to combine multidimensional similarities.Firstly,two concepts from two ontologies are multi-dimensional measured by string-based,semantic-based and structure-based similarities.Then,self-attention model is used to combine above similarity calculations to judge whether the two concepts are similar or not and align them.Experiments on public datasets show that,compared with existing ontology alignment methods,the proposed method can obtain better alignment results by aggregating multi-dimensional similarity features.

Multi-level Inheritance Influence Calculation and Generalization Based on Knowledge Graph

KONG Shi-ming, FENG Yong, ZHANG Jia-yun

Computer Science. 2022, 49 (9): 221-227. doi:10.11896/jsjkx.210700144

Abstract

PDF(2252KB) ( 555 )

References | Related Articles | Metrics

Influence calculation and analysis are widely used in social networks,web page importance evaluation and other fields.There is still a lack of effective and universal solution for the multi-level influence calculation with inheritance chain and time span factors.At the same time,the calculation of maximizing the propagation influence is an NP hard problem,whose approximate algorithm has low accuracy and complicated computation.In order to solve the above problems,this paper proposes a multi-level inheritance influence and generalization algorithm based on knowledge graph to realize the calculation of inheritance influence and inheritance relationship.The algorithm uses the breadth first search hierarchy computing model of knowledge graph,and takes into account the time span constraints to calculate the inheritance influence and inheritance chain.In order to optimize the computational efficiency,the strategy of depth first search and different levels with different weights is further used to only calculate the influence of the top n levels.The above method can not only calculate the inheritance influence and inheritance chain well,but also can be generalized into various communication influence calculation models.On this basis,this paper proposes a local optimal search similarity algorithm to maximize the propagation influence by selecting the nodes with large propagation influence as spare nodes.It achieves competitive results in running speed and the maximum number of propagation nodes.Finally,the effectiveness of the proposed method is verified by a variety of simulation experiments.

Foundation of Indoor Information Services:A Survey of Low-cost Localization Techniques

SHAO Zi-hao, YANG Shi-yu, MA Guo-jie

Computer Science. 2022, 49 (9): 228-235. doi:10.11896/jsjkx.210900260

Abstract

PDF(1688KB) ( 858 )

References | Related Articles | Metrics

Recently,with the development of Internet of things technology and the proposal of smart city concept,location-based service is developing rapidly,especially,the outdoor location services provided by global positioning system (GPS) based on satellite signals have penetrated into every aspect of out daily life.However,GPS is not applicable for indoor space due to the low localization accuracy,affected by the complex indoor environment.In order to improve the localization accuracy,indoor localization techniques are proposed.The techniques which utilize the existing devices such as Wi-Fi,low energy Bluetooth (BLE) are attracting more and more attention due to their advantages of low cost and easy deployment.This paper surveys the recent research work of low-cost indoor localization techniques with the basic motivations,implementations and their localization performance.Finally,the future development trend is prospected.

Construction and Distribution Method of REM Based on Edge Intelligence

LIU Xing-guang, ZHOU Li, LIU Yan, ZHANG Xiao-ying, TAN Xiang, WEI Ji-bo

Computer Science. 2022, 49 (9): 236-241. doi:10.11896/jsjkx.220400148

Abstract

PDF(2114KB) ( 977 )

References | Related Articles | Metrics

Radio environment map(REM) can assist cognitive users to accurately perceive and utilize spectrum holes,achieve interference coordination between network nodes,and improve the spectrum efficiency and robustness of wireless networks.However,when cognitive users utilize and share REM,there are problems of high computational complexity and high distribution delay overhead,which limit cognitive users' ability to perceive spatial spectrum situation in real time.To solve this problem,this paper proposes a reinforcement learning-based REM construction and distribution method in mobile edge intelligence networks.First,we employ a low-complexity construction technique that combines kriging interpolation and super-resolution for REM construction.Second,we model the computational offload strategy selection problem during REM construction and distribution as a mixed-integer nonlinear programming problem by using edge computing.Finally,we combine artificial intelligence technology and edge computing technology,and propose a centralized training,distributed execution reinforcement learning framework to learn REM construction and distribution strategies in different network scenarios.Simulation results show that the proposed method has good adaptability,and it can effectively reduce the energy consumption and delay of REM construction and distribution,and support the near real-time application of REM by cognitive users in mobile edge network scenarios.

Dynamic Pricing-based Vehicle Collaborative Computation Offloading Scheme in VEC

SUN Hui-ting, FAN Yan-fang, MA Meng-xiao, CHEN Ruo-yu, CAI Ying

Computer Science. 2022, 49 (9): 242-248. doi:10.11896/jsjkx.210700166

Abstract

PDF(2340KB) ( 737 )

References | Related Articles | Metrics

Vehicular edge computing(VEC) is an important application of mobile edge computing(MEC) in Internet of vehicles.In VEC,to meet the computing requirement of task vehicles(TaV),TaV can pay to offload tasks to the VEC server or service vehicles(SeV)with abundant idle computing resources.For the VEC provider,one of its goals is to maximize revenue.Since the computing requirements and the computing resources of the system change dynamically,it is an important issue to design a reasonable pricing strategy in vehicle collaboration scenarios.To solve this problem,this paper designs a dynamic pricing strategy.In this strategy,service prices of the VEC server and SeV are adjusted dynamically according to the relationship of the supply and demand of computing resources.On this basis,a vehicle collaborative computation offloading scheme is designed to maximize provider's revenue.By transforming the revenue maximization problem of VEC provider under the delay constraint into a multi-user matching problem,the offloading results are obtained using the Kuhn-Munkres(KM)algorithm.Simulation results show that compared to existing strategies,with this dynamic pricing strategy,the prices of the VEC server and SeV can be adjusted dynamically with the supply and demand of resources,so as to maximize the provider's revenue.Compared to existing offloading schemes,this scheme can improve the provider's revenue while meeting task delay.

Functional Architecture to Intelligent Computing Power Network

HU Yu-jiao, JIA Qing-min, SUN Qing-shuang, XIE Ren-chao, HUANG Tao

Computer Science. 2022, 49 (9): 249-259. doi:10.11896/jsjkx.220500222

Abstract

PDF(4066KB) ( 1114 )

References | Related Articles | Metrics

The computing power network as a new research fieldis in urgent need to improve intelligence and provide on-demand services.To solve the problems,an intelligent computing power network that combines cloud-edge-terminal computing resource,communication resource,and AI approaches together is proposed.At the same time,the system-level intelligence would be built from two aspects of endogenous intelligence and application intelligence.The endogenous intelligence refers to the abilities of self-perception,self-adaptation,self-decision and self-learning to make the computing power network ensure the accurate operation of the system.The application intelligence refers to the abilities in resource deployment,business arrangement and cognition,so that the computing power network could enhance the adaptability of the businesses.Further,a functional architecture is constructed,which would gradually build the endogenous intelligence and application intelligence in the layers of basic resource,resource mana-gement,business choreography,operation service and optimization.Finally,three important scenarios are selected from two domains,i.e.workshop logistics and quality control based on machine vision in the smart manufacturing,road and community monitoring in the intelligent security.Three groups of simulations are designed based on the scenarios,respectively.Experimental results show that while adopting the intelligent computing power network with endogenous intelligent and application intelligence,in the scenario of workshop logistics,the performance improvement is related with the scale of the scenarios,the planning time could be improved by approximately 2～50 times and the planning result could be improved by about 2～5 times.In the scenario of quality control based on machine vision,the deploymentcost of computing equipment would reduce to 1/5 of the original and the detection accuracy could improve by about 4.5%.In the scenario of road and community monitoring,the deployment cost of computing equipment could reduceto 1/10 of the original.

Collaborative Multicast Proactive Caching Scheme Based on AAE

LIU Xin, WANG Jun, SONG Qiao-feng, LIU Jia-hao

Computer Science. 2022, 49 (9): 260-267. doi:10.11896/jsjkx.210800019

Abstract

PDF(2941KB) ( 592 )

References | Related Articles | Metrics

With the increasing number of user terminals and the development of 5G technology,a network has been formed where macro base stations and small base stations co-exist.Meanwhile,applications such as ultra-high resolution video and cloud VR/AR have higher requirements for latency.In order to reduce the latency in 5G networks,a cooperative multicast proactive caching scheme based on adversarial automatic coding is proposed in this paper.In this scheme,firstly,users are divided into different groups based on their characteristics.And then the content that the group may request will be predicated by using AAE.To reduce the redundancy of cached contents,the ant colony algorithm is used to pre-deploy the predicted contents to each small base station.Finally,in the content distribution phase,if a user requests a content with high popularity,the content will be proactively cached in a multicast manner to other users in this group who don't send the request,otherwise it is distributed in a normal manner.Simulation results show that the CMPCAAE scheme outperforms the classical caching scheme in terms of average delay and missing ratio of the system.

Study on Wireless Communication Network Architecture and Access Control Algorithm in Aircraft

GUO Peng-jun, ZHANG Jing-zhou, YANG Yuan-fan, YANG Shen-xiang

Computer Science. 2022, 49 (9): 268-274. doi:10.11896/jsjkx.210700220

Abstract

PDF(3716KB) ( 707 )

References | Related Articles | Metrics

With the rapid development of avionics system,a large number of devices and sensors are connected to the in-plane network,which makes the architecture of the in-plane communication network complex and heavy.Wireless communication network can effectively solve many problems,such as complicated wiring,heavy weight and difficult line fault detection.However,wireless network still has certain limitations in real-time and reliability,which is precisely the most concerned problem of airborne interconnection system.In this paper,the existing wired communication network architecture is analyzed,a hybrid communication network architecture of wireless access network and wired backbone network is designed according to its communication characteristics.The candidate wireless communication schemes are evaluated and selected.The fixed time slot of wireless network access is improved to dynamic allocation according to traffic,the mathematical model of dynamic time slot allocation is established,and the TDMA period and the optimal time slot allocation strategy are designed.Finally,a typical airborne network task shows that the strategy can improve the network utilization rate from 36.5% to 41.8% on the premise of ensuring that the system is schedulable,which verifies the effectiveness of the method.

Incentive Mechanism Based on Multi-constrained Worker Selection in Mobile Crowdsourcing

FU Yan-ming, ZHU Jie-fu, JIANG Kan, HUANG Bao-hua, MENG Qing-wen, ZHOU Xing

Computer Science. 2022, 49 (9): 275-282. doi:10.11896/jsjkx.210700129

Abstract

PDF(3356KB) ( 593 )

References | Related Articles | Metrics

With the rapid development of mobile crowdsourcing,crowdsourcing programs in the market have sprung up.They distribute tasks and use the power of the crowd to perform the tasks for collecting data and an effective incentive mechanism in mobile crowdsourcing becomes very important.However,the existing incentive mechanisms nowadays partially consider the reputation value,location and execution time of workers,which makes it difficult for crowdsourcing platform to select high-quality workers and assign multiple tasks on limited budgets or other constraints.To solve the above problems,this paper proposes an incentive mechanism on the basis of the multi-constrained worker selection (MSIM),which relies on two related algorithms.One is the algorithm of worker selection based on improved reverse auction model,which comprehensively considers many important limitations to select great workers to perform the tasks,such as worker reputation,geographical location,task completion degree and result quality.The other is the algorithm of reward and punishment by evaluation,which contains the evaluation of task-perceiving results and workers' reputation.The experimental results showed that not only can MSIM select excellent workers,but also it improved the credibility of the task results and the reputation of workers.It is proved within this paper that the MSIM is an effective incentive mechanism.

PDR Indoor Positioning Method Based on M2M Encounter Region

TANG Qing-hua, WANG Mei, TANG Chao-chen, LIU Xin, LIANG Wen

Computer Science. 2022, 49 (9): 283-287. doi:10.11896/jsjkx.210800270

Abstract

PDF(2376KB) ( 508 )

References | Related Articles | Metrics

In indoor positioning,the main advantage of pedestrian dead reckoning(PDR)is that the user only needs to have a smart phone to realize positioning,without relying on the external environment.However,there is a large cumulative error.Ge-nerally,it is necessary to combine Bluetooth,WiFi,geomagnetic or other technologies to improve the positioning accuracy.How-ever, this method requires some hardware nodes and a fingerprint database to be built for this purpose.To solve this problem,an indoor positioning method based on correcting PDR in machine to machine(M2M)area is proposed.Firstly,a distance measurement area is set up during pedestrian travel.Secondly,the distance between pedestrian mobile phones and other mobile phones is measured in this region.Finally,the positioning error and accuracy of PDR are corrected by trilateral positioning method.The method has the advantages that no additional hardware facilities are required.Experimental results show that,compared with the traditional PDR positioning,this method is suitable for long-time positioning,and the average positioning error is reduced to 0.36 m,with high positioning accuracy.

Research Progress and Analysis on Intelligent Cryptology

NING Han-yang, MA Miao, YANG Bo, LIU Shi-chang

Computer Science. 2022, 49 (9): 288-296. doi:10.11896/jsjkx.220300053

Abstract

PDF(1932KB) ( 2075 )

References | Related Articles | Metrics

The rapid development of artificial intelligence and 5G network technology has opened a new era of interconnection of all things.The great improvement of computing power has threatened the traditional cryptographic algorithm based on the theory of computational difficulty.Data security and communication security have become key problems to be solved urgently in the era of Internet of things,hence cryptology has entered an intelligence era.The new generation of intelligent cryptology mainly consists of two core technologies:intelligent cryptographic algorithm based on neural network and intelligent cryptanalysis based on machine learning.The former uses the nonlinear characteristics of neural network to design the encryption process and improve the security of ciphertext.The latter trains the machine learning model through the clear ciphertext set to obtain the ciphertext features and improve the ciphertext decoding efficiency.This paper briefly reviews the development of cryptographic algorithms,discusses machine learning methods on intelligent cryptology,focuses on combing the latest progress of cryptographic algorithms and cryptanalysis intelligence at home and abroad,analyzes the advantages and disadvantages of intelligent cryptology at present,and discusses the research direction and challenges in the future.

Federated Learning Scheme Based on Secure Multi-party Computation and Differential Privacy

TANG Ling-tao, WANG Di, ZHANG Lu-fei, LIU Sheng-yun

Computer Science. 2022, 49 (9): 297-305. doi:10.11896/jsjkx.210800108

Abstract

PDF(2783KB) ( 1953 )

References | Related Articles | Metrics

Federated learning provides a novel solution to collaborative learning among untrusted entities. Through a local-trai-ning-and-central-aggregation pattern,the federated learning algorithm trains a global model while protects local data privacy of each entity. However,recent studies show that local models uploaded by clients and global models produced by the server may still leak users' private information. Secure multi-party computation and differential privacy are two mainstream privacy-preserving techniques,which are used to protect the privacy of computation process and computation outputs respectively. There are few works that exploit the benefits of these two techniques at the same time. This paper proposes a privacy-preserving federated learning scheme for deep learning by combining secure multi-party computation and differential privacy. Clients add noise to local models,and secret share them to multiple servers. Servers aggregate these model shares by secure multi-party computation to obtain a private global model. The proposed scheme not only protects the privacy of local model updates uploaded by clients,but also prevents adversaries from inferring sensitive information from globally shared data such as aggregated models. The scheme also allows dropout of unstable clients and is compatible with complex aggregation functions. In addition,it can be naturally extended to the decentralized setting for real-world applications where no trusted centers exist. We implement our system in Python and Pytorch. Experiments validate that the proposed scheme achieves the same level of efficiency and accuracy as plaintext fede-rated learning.

Network Security Risk Assessment Framework Based on Tactical Correlation

LIU Jie-ling, LING Xiao-bo, ZHANG Lei, WANG Bo, WANG Zhi-liang, LI Zi-mu, ZHANG Hui, YANG Jia-hai, WU Cheng-nan

Computer Science. 2022, 49 (9): 306-311. doi:10.11896/jsjkx.210600171

Abstract

PDF(2510KB) ( 816 )

References | Related Articles | Metrics

Power system network is one of the important targets of cyber attack.In order to ensure the safe operation of power system,network managers need to evaluate the network security risk.Usually,existing network security risk assessment framework only aims at a single scenario,and can not find the strategic attackers who use a variety of low-risk methods to achieve high-risk threat targets from large quantities of network security alerts.In order to meet the above challenges,this paper proposes a network security risk assessment method based on tactical correlation.In this method,the warning information generated on va-rious network security detection devices when an attacker implements a multi-step attack is associated to form an attack chain,and the security risk of the organization intranet is evaluated by calculating the threat,vulnerability,impact score of each node in the attack chain and the risk score of the whole attack chain.In order to verify the effectiveness and robustness of the proposed method,this paper selects a representative example to illustrate the specific implementation process of the proposed method for network security risk assessment in the organizational intranet.The example shows that the network security risk assessment framework based on the tactical association can correctly assess the harm of multi-step attack caused by low-risk alarm association to achieve high-risk targets,and is more robust than the traditional single scenario analysis method,which can better provide decision-making basis for organization decision-makers in network security risk management.

Research and Implementation of Parallel Method in Blockchain and Smart Contract

WANG Zi-kai, ZHU Jian, ZHANG Bo-jun, HU Kai

Computer Science. 2022, 49 (9): 312-317. doi:10.11896/jsjkx.210800102

Abstract

PDF(3329KB) ( 1204 )

References | Related Articles | Metrics

With the continuous maturity of blockchain technology,there are more and more blockchain applications for enterprises that can provide a safe,anonymous and non-tamperable transaction environment.Traditional blockchain architecture is faced with problems such as low performance and insufficient scalability.It can neither meet the needs of high concurrency nor the big data application scenarios for enterprise-level applications.In order to better adapt to the more abundant application scenarios and give full paly to the value of blockchain technology,this paper proposes a simple practical byzantine fault tolerance(SBFT)consensus algorithm to improve the efficiency of the consensus phase,and a Task parallel smart contract model is proposed to make full use of the parallelism efficiency of multi-core systems.we have improved the traditional blockchain system architecture to have the characteristics of light weight,low coupling,and smart contract scalability,which facilitates the secondary development of enterprise applications.On this basis,the ParaChain blockchain and smart contract system are developed.Experimental results show that the performance and scalability of the ParaChain blockchain based on parallelization technology is greatly improved compared to the blockchain system based on the traditional PBFT consensus protocol.

Privacy-preserving Linear Regression Scheme and Its Application

LYU You, WU Wen-yuan

Computer Science. 2022, 49 (9): 318-325. doi:10.11896/jsjkx.220300190

Abstract

PDF(1787KB) ( 826 )

References | Related Articles | Metrics

Linear regression is an important and widely used machine learning algorithm.The training of linear regression model usually depends on a large amount of data.In reality,the data set is generally held by different users and contains their privacy information.When multiple users want to gather more data to train a better model,it inevitably involves users' privacy.As a privacy protection technology,homomorphic encryption can effectively solve the problem of privacy leakage in computing.A new privacy preserving linear regression scheme based on hybrid iterative method is designed for the scenario where data sets are distri-buted horizontally on two users.The scheme is divided into two stages.The first stage implements the statistic gradient descent algorithm in the ciphertext domain.In the second stage,a secure two-party fast descent protocol is designed.The core idea of the protocol is based on Jacobi iterative method,which can effectively make up for the poor convergence effect of gradient descent method in practical application,accelerate the convergence of the model,and protect the data privacy of two users while effectively training the linear regression model.The efficiency,communication loss and security of the scheme are analyzed.The scheme is implemented by using C+＋and applied to real data sets.A large number of experimental results show that the scheme can effectively solve the linear regression problem with large scale features.The relative error of decision coefficient is less than 0.001,which show that the application effect of the privacy preserving linear regression model in real data set is close to that obtained directly from unencrypted data,and the scheme can meet the practical application requirements in specific scenarios.

Strcmp-like Function Identification Method Based on Data Flow Feature Matching

HU An-xiang, YIN Xiao-kang, ZHU Xiao-ya, LIU Sheng-li

Computer Science. 2022, 49 (9): 326-332. doi:10.11896/jsjkx.220200163

Abstract

PDF(2190KB) ( 785 )

References | Related Articles | Metrics

Embedded devices have become visible everywhere,and they are used in a range of security-critical and privacy-sensitive applications.However,recent studies show that many embedded devices have backdoor,of which hard-coded backdoor(password backdoor) is the most common.In the triggering process of password backdoor,strcmp-like functions are necessary and important absolutely.However,the current identification of strcmp-like functions mainly relies on function signature and control flow feature matching.The former can't recognize user-defined strcmp-like functions,and the identify effect is greatly affected by the compile environment.The latter has high false positive rate and false negative rate.To solve the above problems,this paper proposes a novel strcmp-like recognition technology CMPSeek.This method builds a model for strcmp-like function identification based on the analysis of control flow and data flow characteristics,which is used to identify strcmp-like functions in binary programs,and is suitable for stripped binary programs.Furthermore,ARM,MIPS,PPC and x86/64 instruction sets are supported by converting binary codes to the intermediate language representation VEX IR codes.Experimental results show that CMPSeek has better results in accuracy rate and recall rate than FLIRT and SaTC in the absence of source code,function name and other information.

Belief Driven Attack and Defense Policy Optimization Mechanism in Honeypot Game

JIANG Yang-yang, SONG Li-hua, XING Chang-you, ZHANG Guo-min, ZENG Qing-wei

Computer Science. 2022, 49 (9): 333-339. doi:10.11896/jsjkx.220400011

Abstract

PDF(2093KB) ( 625 )

References | Related Articles | Metrics

As a typical deception defense means,honeypot technology is of great significance in actively trapping attackers.The existing design methods mainly optimize the trapping decision of honeypot through the game model,ignoring the impact of the attacker's belief on the game decision of both sides.There are some shortcomings,such as weak adaptive optimization decision-making ability,easy to be seen through and used by the attacker and so on.Therefore,a belief based honeypot game mechanism(BHGM) is proposed.Based on the multi round game process of attacker completing the task,BHGM focuses on the impact of honeypot action on attacker's belief and the impact of belief on whether the attacker continues to attack.At the same time,a belief driven algorithm for solving the optimal attack and defense strategy is designed based on the upper confidence bound apply to tree(UCT).Simulation results show that the belief driven attacker strategy can choose to continue the attack or stop the loss in time based on the current belief to obtain the maximum profit,while the belief driven honeypot strategy can reduce attacker's suspicion as much as possible to lure him to continue the attack and obtain greater profit.

Bi-histogram Shifting Reversible Data Hiding Method After Compressed Differences

HAO Jie, PING Ping, FU De-yin, ZHAO Hong-ze

Computer Science. 2022, 49 (9): 340-346. doi:10.11896/jsjkx.220300238

Abstract

PDF(2757KB) ( 578 )

References | Related Articles | Metrics

Reversible data hiding(RDH) based on histogram shifting(HS) is the most common technology in current information hiding,especially for the combined method of difference extension and histogram shifting,which can achieve high embedding capacity and low image distortion.In this paper,a reversible information hiding method of bi-histogram shifting after compressed difference is proposed.The algorithm improves the defect of insufficient embedding capacity of the existing method based on histogram shifting by synthesizing three methods of compression,difference and optimized histogram shifting.At the same time,the processing method for the overflow of the image pixel value in the shifting process is also given.At the receiving end,not only can the data be completely extracted,but also lossless image recovery can be performed.After the experiment,a comparison is made with the four current popular schemes.Our method outperforms existing histogram shifting based algorithms in terms of embedding capacity.Compared with other methods,its embedded capacity improves by 23%,11%,57% and 93%.Experimental results show that the proposed method greatly increases the embedding capacity and can effectively realize reversible information hiding with large embedding capacity.

LBS Mobile Privacy Protection Scheme Based on Random Onion Routing

WANG Lei, LI Xiao-yu

Computer Science. 2022, 49 (9): 347-354. doi:10.11896/jsjkx.210800077

Abstract

PDF(1953KB) ( 600 )

References | Related Articles | Metrics

In order to ensure the location privacy of mobile nodes when using location-based services,a mobile privacy protection scheme for location based service(LBS)based on random onion routing is proposed.This scheme integrates random onion routing and hybrid encryption to ensure the privacy of mobile node location and the security of query requests.The mobile node randomly selects several nodes in the network before sending the query request to the LBS server to build an onion path,and forwards the query request along this path in turn until the LBS server receives the message.Then,the LBS sends the query result to the sen-ding node along the reverse direction of the onion path.In order to realize the anonymity of the sending node,the addresses of each layer on the randomly constructed onion path are encrypted with a combination of symmetric encryption and asymmetric encryption.In this way,layers of encryption are used to generate the final onion path.Each hop node in the path can get only its next-hop node.Neither the LBS server nor any relay node in the onion routing path can know which node is the sending node,so that the location privacy of the sending node can be kept.On the other hand,in order to ensure that no third party can get the query request or query result,the sending node first uses a symmetric key to encrypt the query request.Then it uses the public key of the LBS server to encrypt the symmetric key.Finally it attaches the symmetric key ciphertext to the query request ciphertext and send it to the LBS server.The LBS server will also return the encrypted query result.Experimental results show that the average response time of this scheme increases slowly with the increase of the number of nodes in the system.The average response time will not increases sharply with the increase of the number of nodes,which will lead to system paralysis.So the system has good stability and scalability.The onion path is randomly selected which does not depend on a specific node,so the scheme has better robustness.

Privacy-preserving Hamming and Edit Distance Computation and Applications

DOU Jia-wei

Computer Science. 2022, 49 (9): 355-360. doi:10.11896/jsjkx.220100241

Abstract

PDF(1484KB) ( 1210 )

References | Related Articles | Metrics

With the rapid development of information technology,privacy-preserving multiparty cooperative computation is becoming more and more popular.Secure multiparty computation is a key technology to address such problems.In scientific research and practical applications,people measure the similarity of two strings with Hamming(edit) distance.It is of great significance to study privacy-preserving Hamming(edit) distance computation.This paper studies privacy-preserving Hamming(edit) distance computation.First,we transform Hamming distance computation to inner product computation of vectors,and then use Okamoto-Uchiyama(OU) cryptosystem and encryption-and-choose technique to design protocol for Hamming distance.Second,we give each alphabet of a string a number,transform edit distance to determine whether the difference of the number of two alphabets is 0,and use OU cryptosystem to design a privacy-preserving edit distance computation protocol.The security of the protocol is strictly proved,the computational complexity of the protocol is analyzed,the actual implementation efficiency of the protocol is tested and compared with the existing results.Theoretical analysis and experimental results show that our protocols are efficient.