Skip to main content
main-content

Über dieses Buch

The two-volume set of LNCS 10941 and 10942 constitutes the proceedings of the 9th International Conference on Advances in Swarm Intelligence, ICSI 2018, held in Shanghai, China, in June 2018. The total of 113 papers presented in these volumes was carefully reviewed and selected from 197 submissions. The papers were organized in topical sections namely: multi-agent systems; swarm robotics; fuzzy logic approaches; planning and routing problems; recommendation in social media; predication; classification; finding patterns; image enhancement; deep learning; theories and models of swarm intelligence; ant colony optimization; particle swarm optimization; artificial bee colony algorithms; genetic algorithms; differential evolution; fireworks algorithm; bacterial foraging optimization; artificial immune system; hydrologic cycle optimization; other swarm-based optimization algorithms; hybrid optimization algorithms; multi-objective optimization; large-scale global optimization.

Inhaltsverzeichnis

Frontmatter

Multi-agent Systems

Frontmatter

Path Following of Autonomous Agents Under the Effect of Noise

In this paper, we adopt the architecture of the Lyapunov-based Control Scheme (LbCS) and integrate a leader-follower approach to propose a collision-free path following strategy of a group of mobile car-like robots. A robot is assigned the responsibility of a leader, while the follower robots position themselves relative to the leader so that the path of the leader robot is followed with arbitrary desired clearance by the follower robot, avoiding any inter-robot collision while navigating in a terrain with obstacles under the influence of noise. A set of artificial potential field functions is proposed using the control scheme for the avoidance of obstacles and attraction to their designated targets. The effectiveness of the proposed nonlinear acceleration control laws is demonstrated through computer simulations which prove the efficiency of the control technique and also demonstrates its scalability for larger groups.

Krishna Raghuwaiya, Bibhya Sharma, Jito Vanualailai, Parma Nand

Development of Adaptive Force-Following Impedance Control for Interactive Robot

This paper presented a safety approach for the interactive manipulator. At first, the basic compliance control of the manipulator is realized by using the Cartesian impedance control, which inter-related the external force and the end position. In this way, the manipulator could work as an external force sensor. A novel force-limited trajectory was then generated in a high dynamics interactive manner, keeping the interaction force within acceptable tolerance. The proposed approach also proved that the manipulator was able to contact the environment compliantly, and reduce the instantaneous impact when collision occurs. Furthermore, adaptive dynamics joint controller was extended to all the joints for complementing the biggish friction. Experiments were performed on a 5-DOF flexible joint manipulator. The experiment results of taping the obstacle, illustrate that the interactive robot could keep the desired path precisely in free space, and follow the demand force in good condition.

Huang Jianbin, Li Zhi, Liu Hong

A Space Tendon-Driven Continuum Robot

In order to avoid the collision of space manipulation, a space continuum robot with passive structural flexibility is proposed. This robot is composed of two continuum joints with elastic backbone and driving tendons made of NiTi alloy. The kinematic mapping and the Jacobian matrix are obtained through the kinematic analysis. Moreover, an inverse kinematics based closed-loop controller is designed to achieve position tracking. Finally, a simulation and an experiment is carried out to validate the workspace and control algorithm respectively. The results show that this robot can follow a given trajectory with satisfactory accuracy.

Shineng Geng, Youyu Wang, Cong Wang, Rongjie Kang

A Real-Time Multiagent Strategy Learning Environment and Experimental Framework

Many problems in the real world can be attributed to the problem of multiagent. The study on the issue of multiagent is of great significance to solve these social problems. This paper reviews the research on multiagent based real-time strategy game environments, and introduces the multiagent learning environment and related resources. We choose a deep learning environment based on the StarCraft game as a research environment for multiagent collaboration and decision-making, and form a research mentality focusing mainly on reinforcement learning. On this basis, we design a verification platform for the related theoretical research results and finally form a set of multiagent research system from the theoretical method to the actual platform verification. Our research system has reference value for multiagent related research.

Hongda Zhang, Decai Li, Liying Yang, Feng Gu, Yuqing He

Transaction Flows in Multi-agent Swarm Systems

The article presents a mathematical model of transaction flows between individual intelligent agents in swarm systems. Assuming that transaction flows are Poisson ones, the approach is proposed to the analytical modeling of such systems. Methods for estimating the degree of approximation of real transaction flows to Poisson flows based on Pearson’s criterion, regression, correlation and parametric criteria are proposed. Estimates of the computational complexity of determining the parameters of transaction flows by using the specified criteria are shown. The new criterion based on waiting functions is proposed, which allows obtaining a good degree of approximation of an investigated flow to Poisson flow with minimal costs of computing resources. That allows optimizing the information exchange processes between individual units of swarm intelligent systems.

Eugene Larkin, Alexey Ivutin, Alexander Novikov, Anna Troshina

Event-Triggered Communication Mechanism for Distributed Flocking Control of Nonholonomic Multi-agent System

As the scale of multi-agent systems (MAS) increases, communication becomes a bottleneck. In this paper, we propose an event-triggered mechanism to reduce the inter-agent communication cost for the distributed control of MAS. Communication of an agent with others only occurs when event triggering condition (ETC) is met. In the absence of communication, other agents adopt an estimation process to acquire the required information about the agent. Each agent has an above estimation process for itself and another estimation based on Kalman Filter, the latter can represent its actual state considering the measurement value and error from sensors. The error between the two estimators indicates whether the estimator in other agents can maintain a relatively accurate state estimation for this agent, and decides whether the communication is triggered. Simulations demonstrate the effectiveness and advantages of the proposed method for the distributed control of flocking in both Matlab and Gazebo.

Weiwei Xun, Wei Yi, Xi Liu, Xiaodong Yi, Yanzhen Wang

Deep Regression Models for Local Interaction in Multi-agent Robot Tasks

A direct data-driven path planner for small autonomous robots is a desirable feature of robot swarms that would allow each agent of the system to directly produce control actions from sensor readings. This feature allows to bring the artificial system closer to its biological model, and facilitates the programming of tasks at the swarm system level. To develop this feature it is necessary to generate behavior models for different possible events during navigation. In this paper we propose to develop these models using deep regression. In accordance with the dependence of distance on obstacles in the environment along the sensor array, we propose the use of a recurrent neural network. The models are developed for different types of obstacles, free spaces and other robots. The scheme was successfully tested by simulation and on real robots for simple grouping tasks in unknown environments.

Fredy Martínez, Cristian Penagos, Luis Pacheco

Multi-drone Framework for Cooperative Deployment of Dynamic Wireless Sensor Networks

A system implementing a proposed framework for using multiple-cooperating-drones in the deployment of a dynamic sensor network is completed and preliminary tests performed. The main components of the system are implemented using a genetic strategy to create the main elements of the framework. These elements are sensor network topology, a multi objective genetic algorithm for path planning, and a cooperative coevolving genetic strategy for solving the optimal cooperation problem between drones. The framework allows for mission re-planning with changes to drone fleet status and environmental changes as a part of making a fully autonomous system of drones.

Jon-Vegard Sørli, Olaf Hallan Graven

Swarm Robotics

Frontmatter

Distributed Decision Making and Control for Cooperative Transportation Using Mobile Robots

This paper introduces a distributed control scheme tailor-made to the task of letting a swarm of mobile robots push an object through a planar environment. Crucially, there is no centralized control instance or inter-robot hierarchy, and therefore, all decisions are made in a distributed manner. For being able to cooperate, the robots communicate, although the communication sampling time may be several times longer than the control sampling time. Most characteristic for the approach, distributed model predictive controllers are used to achieve a smooth transportation performance with the predicted control errors utilized to plan a suitable object trajectory. Challenging simulation scenarios show the applicability of the approach to the transportation task.

Henrik Ebel, Peter Eberhard

Deep-Sarsa Based Multi-UAV Path Planning and Obstacle Avoidance in a Dynamic Environment

This study presents a Deep-Sarsa based path planning and obstacle avoidance method for unmanned aerial vehicles (UAVs). Deep-Sarsa is an on-policy reinforcement learning approach, which gains information and rewards from the environment and helps UAV to avoid moving obstacles as well as finds a path to a target based on a deep neural network. It has a significant advantage over dynamic environment compared to other algorithms. In this paper, a Deep-Sarsa model is trained in a grid environment and then deployed in an environment in ROS-Gazebo for UAVs. The experimental results show that the trained Deep-Sarsa model can guide the UAVs to the target without any collisions. This is the first time that Deep-Sarsa has been developed to achieve autonomous path planning and obstacle avoidance of UAVs in a dynamic environment.

Wei Luo, Qirong Tang, Changhong Fu, Peter Eberhard

Cooperative Search Strategies of Multiple UAVs Based on Clustering Using Minimum Spanning Tree

Rate of revenue (ROR) is significant for unmanned aerial vehicle (UAV) to search targets located in probabilistic positions. To improve search efficiency in a situation of multiple static targets, this paper first transfers a continuous area to a discrete space by grid division and proposes some related indexes in the UAV search issue. Then, cooperative strategies of multiple UAVs are studied in the searching process: clustering partition of search area based on minimum spanning tree (MST) theory is put forward as well as path optimization using spiral flying model. Finally, a series of simulation experiments are carried out through the method in this paper and two compared algorithms. Results show that: optimized cooperative strategies can achieve greater total revenue and more stable performance than the other two.

Tao Zhu, Weixiong He, Haifeng Ling, Zhanliang Zhang

Learning Based Target Following Control for Underwater Vehicles

Target following of underwater vehicles has attracted increasingly attentions on their potential applications in oceanic resources exploration and engineering development. However, underwater vehicles confront with more complicated and extensive difficulties in target following than those on the land. This study proposes a novel learning based target following control approach through the integration of type-II fuzzy system and support vector machine (SVM). The type-II fuzzy system allows researchers to model and minimize the effects of uncertainties of changing environment in the rule-based systems. In order to improve the vehicle capacity of self-learning, an SVM based learning approach has been developed. Through genetic algorithm generating and mutating fuzzy rules candidate, SVM learning and optimization, one can obtain optimized fuzzy rules. Tank experiments have been performed to verify the proposed controller.

Zhou Hao, Huang Hai, Zhou Zexing

Optimal Shape Design of an Autonomous Underwater Vehicle Based on Gene Expression Programming

A novel strategy combining gene expression programming and crowding distance based multi-objective particle swarm algorithm is presented in this paper to optimize an underwater robot’s shape. The gene expression programming method is used to establish the surrogate model of resistance and surrounded volume of the robot. After that, the resistance and surrounded volume are set as two optimized factors and Pareto optimal solutions are then obtained by using multi-objective particle swarm optimization. Finally, results are compared with the hydrodynamic calculations. Result shows the efficiency of the method proposed in the paper in the optimal shape design of an underwater robot.

Qirong Tang, Yinghao Li, Zhenqiang Deng, Di Chen, Ruiqin Guo, Hai Huang

GLANS: GIS Based Large-Scale Autonomous Navigation System

The simultaneous localization and mapping (SLAM) systems are widely used for self-localization of a robot, which is the basis of autonomous navigation. However, the state-of-art SLAM systems cannot suffice when navigating in large-scale environments due to memory limit and localization errors. In this paper, we propose a Geographic Information System (GIS) based autonomous navigation system (GLANS). In GLANS, a topological path is suggested by GIS database and a robot can move accordingly while being able to detect the obstacles and adjust the path. Moreover, the mapping results can be shared among multi-robots to re-localize a robot in the same area without GPS assistance. It has been proved functioning well in the simulation environment of a campus scenario.

Manhui Sun, Shaowu Yang, Henzhu Liu

Fuzzy Logic Approaches

Frontmatter

Extraction of Knowledge with Population-Based Metaheuristics Fuzzy Rules Applied to Credit Risk

One of the goals of financial institutions is to reduce credit risk. Consequently they must properly select customers. There are a variety of methodologies for credit scoring, which analyzes a wide variety of personal and financial variables of the potential client. These variables are heterogeneous making that their analysis is long and tedious. This paper presents an alternative method that, based on the subject information, offers a set of classification rules with three main characteristics: adequate precision, low cardinality and easy interpretation. This is because the antecedent consists of a small number of attributes that can be modeled as fuzzy variables. This feature, together with a reduced set of rules allows obtaining useful patterns to understand the relationships between data, and make the right decisions for the financial institutions. The smaller the number of analyzed variables of the potential customer, the simpler the model will be. In this way, credit officers may give an answer to the loan application in the shorter time, achieving a competitive advantage for the financial institution. The proposed method has been applied to two databases from the UCI repository, and a database from a credit unions cooperative in Ecuador. The results are satisfactory, as highlighted in the conclusions. Some future lines of research are suggested.

Patricia Jimbo Santana, Laura Lanzarini, Aurelio F. Bariviera

Fuzzy Logic Applied to the Performance Evaluation. Honduran Coffee Sector Case

Every day organizations pay more attention to Human Resources Management, because this human factor is preponderant in the results of it. An important policy is the Performance Evaluation (ED), since it allows the control and monitoring of management indicators, both individual and by process. To analyze the results, decision making in many organizations is done in a subjective manner and in consequence it brings serious problems to them. Taking into account this problem, it is decided to design and apply diffuse mathematical procedures and tools to reduce subjectivity and uncertainty in decision-making, creating work algorithms for this policy, which includes multifactorial weights and analysis with measurement indicators that they allow tangible and reliable results. Statistical techniques (ANOVA) are also used to establish relationships between work groups and learn about best practices.

Noel Varela Izquierdo, Omar Bonerge Pineda Lezama, Rafael Gómez Dorta, Amelec Viloria, Ivan Deras, Lissette Hernández-Fernández

Fault Diagnosis on Electrical Distribution Systems Based on Fuzzy Logic

The occurrence of faults in distribution systems has a negative impact on society, and their effects can be reduced by fast and accurate diagnostic systems that allow to identify, locate, and correct the failures. Since the 1990s, fuzzy logic and other artificial intelligence techniques have been implemented to identify faults in distribution systems. The main objective of this paper is to perform fault diagnoses based on fuzzy logic. For conducting the study, the IEEE 34-Node Radial Test Feeder is used. The data was obtained from ATPDraw-based fault simulation on different nodes of the circuit considering three different fault resistance values of 0, 5, and 10 ohms. The fuzzy rules to identify the type of fault are defined using the magnitudes of the phase and neutral currents. All measurements are taken at the substation, and the results show that the proposed technique can perfectly identify and locate the type of failure.

Ramón Perez, Esteban Inga, Alexander Aguila, Carmen Vásquez, Liliana Lima, Amelec Viloria, Maury-Ardila Henry

Planning and Routing Problems

Frontmatter

Using FAHP-VIKOR for Operation Selection in the Flexible Job-Shop Scheduling Problem: A Case Study in Textile Industry

Scheduling of Flexible Job Shop Systems is a combinatorial problem which has been addressed by several heuristics and meta-heuristics. Nevertheless, the operation selection rules of both methods are limited to an ordered variant wherein priority-dispatching rules are not simultaneously deemed in the reported literature. Therefore, this paper presents the application of dispatching algorithm with operation selection based on Fuzzy Analytic Hierarchy Process (FAHP) and VIKOR methods while considering setup times and transfer batches. Dispatching, FAHP, and VIKOR algorithms are first defined. Second, a multi-criteria decision-making model is designed for operation prioritization. Then, FAHP is applied to calculate the criteria weights and overcome the uncertainty of human judgments. Afterwards, VIKOR is used to select the operation with the highest priority. A case study in the textile industry is shown to validate this approach. The results evidenced, compared to the company solution, a reduction of 61.05% in average delay.

Miguel Ortíz-Barrios, Dionicio Neira-Rodado, Genett Jiménez-Delgado, Hugo Hernández-Palma

A Solution Framework Based on Packet Scheduling and Dispatching Rule for Job-Based Scheduling Problems

Job-based scheduling problems have inherent similarities and relations. However, the current researches on these scheduling problems are isolated and lack references. We propose a unified solution framework containing two innovative strategies: the packet scheduling strategy and the greedy dispatching rule. It can increase the diversity of solutions and help in solving the problems with large solution space effectively. In addition, we propose an improved particle swarm optimization (PSO) algorithm with a variable neighborhood local search mechanism and a perturbation strategy. We apply the solution framework combined with the improved PSO to the benchmark instances of different job-based scheduling problems. Our method provides a self-adaptive technique for various job-based scheduling problems, which can promote mutual learning between different areas and provide guidance for practical applications.

Rongrong Zhou, Hui Lu, Jinhua Shi

A Two-Stage Heuristic Approach for a Type of Rotation Assignment Problem

A two-stage heuristic algorithm is proposed for solving a trainee rotation assignment problem in a local school of nursing and its training hospital. At the first stage, the model is reduced to a simplified assignment problem and solved using a random search procedure. At the second stage, a problem-specific operator is designed and employed with a hill climber to further improve solutions. We benchmark our algorithm with instances generated based on the real-life rules. Results show that the proposed algorithm yields high-quality solutions in less computation time for large scale instances when compared with integer linear programming formulation using the commercial solver Cplex.

Ziran Zheng, Xiaoju Gong

An Improved Blind Optimization Algorithm for Hardware/Software Partitioning and Scheduling

Hardware/software partitioning is an important part in the development of complex embedded system. Blind optimization algorithms are suitable to solve the problem when it is combined with task scheduling. To get hardware/software partitioning algorithms with higher performance, this paper improves Shuffled Frog Leaping Algorithm-Earliest Time First (SFLA-ETF) which is a blind optimization algorithm. Under the supervision of the aggregation factor, the improved algorithm named Supervised SFLA-ETF (SSFLA-ETF) used two steps to better balance exploration and exploitation. Experimental results show that compared with SFLA-ETF and other swarm intelligence algorithms, SSFLA-ETF has stronger optimization ability.

Xin Zhao, Tao Zhang, Xinqi An, Long Fan

Interactive Multi-model Target Maneuver Tracking Method Based on the Adaptive Probability Correction

Non-cooperative target tracking is a key technology for complex space missions such as on-orbit service. To improve the tracking performance during the unknown maneuvering phase of the target, two methods including the IMM (interactive multi-model) algorithm based on extended CW equation and the variable IMM algorithm based on CW and extended CW equation are presented. The analysis and simulation results show that the higher the maneuvering index of the target is, the more obvious the advantages of the classical augmented IMM method are. However, the variable dimension IMM method has consistent performance for all the maneuver index interval of the target, and it is relatively suitable for engineering applications due to the lower complexity of algorithm.

Jiadong Ren, Xiaotong Zhang, Jiandang Sun, Qingshuang Zeng

Recommendation in Social Media

Frontmatter

Investigating Deciding Factors of Product Recommendation in Social Media

With the growing popularity of social media, the number of people using social media to communicate and interact with others has increased steadily. As a result, social commerce has become a new phenomenon. In the past, most of the product recommendations in microblogging only dealt with personal preferences and interests, and ignored other possible factors such as Crowd Interest, Popularity of Products, Reputation of Creators, Types of Preference and Recent. Nowadays, these variables used by Facebook to recommend posts to their users. Therefore, this research adapted those five aspects and analyzed their effectiveness to recommend products on social media. This study used the Plurk API to develop and implement recommended robots that recommend products at specific times of the day so that they can get product information and meet recommended tasks in the social circle. The empirical results showed that the Interest, Popularity and Type have significant impacts on recommendation effectiveness.

Jou Yu Chen, Ping Yu Hsu, Ming Shien Cheng, Hong Tsuen Lei, Shih Hsiang Huang, Yen-Huei Ko, Chen Wan Huang

Using the Encoder Embedded Framework of Dimensionality Reduction Based on Multiple Drugs Properties for Drug Recommendation

After obtaining a large amount of drug information, how to extract the most important features from various high-dimensional attribute datasets for drug recommendation has become an important task in the initial stage of drug repositioning. Dimensionality reduction is a necessary and important task for getting the best features in next step. In this paper, three important attribute data about the drugs (i.e., chemical structures, target proteins and side effects) are selected, and two deep frameworks named as F1 and F2 are used to accomplish the task of dimensionality reduction. The processed data are used for recommending new indications by collaborative filtering algorithm. In order to compare the results, two important values of Mean Absolute Error (MAE) and Coverage are selected to evaluate the performance of the two frameworks. Through the comparison with the results of Principal Components Analysis (PCA), it shows that the two deep frameworks proposed in this paper perform better than PCA and can be used for dimensionality reduction task in the future in drug repositioning.

Jun Ma, Ruisheng Zhang, Rongjing Hu, Yong Mu

A Personalized Friend Recommendation Method Combining Network Structure Features and Interaction Information

With the popularity of social network platforms in the crowd, more and more platforms begin to develop friend recommendation services to fit the users’ demands. Current research on friend recommendation strategies are mainly based on the nodes structural characteristics and path information of the friendship network. The recommendation strategies that consider node information are more efficient for large-scale networks, such as the Adamic-Adar Index. However, it solely utilizes the degree information of common neighbors and ignores the structural characteristics of the target nodes themselves. In this paper we attempted to improve the friend recommendation performance by incorporating the structural characteristics of the target nodes and the interactions between these nodes into the Adamic-Adar Index. In order to verify the effectiveness of our proposed algorithms, we conducted several groups of comparative experiments. The experimental results show that our proposed algorithm can effectively improve the recommendation performance comparing with the benchmark.

Chen Yang, Tingting Liu, Lei Liu, Xiaohong Chen, Zhiyong Hao

A Hybrid Movie Recommendation Method Based on Social Similarity and Item Attributes

With the increasing demand for personalized recommendation, traditional collaborative filtering cannot satisfy users’ needs. Social behaviors such as tags, comments and likes are becoming more and more popular among the recommender system users, and are attracting the attentions of the researchers in this domain. The behavior characteristics can be integrated with traditional interest community and some content features. In this paper, we put forward a hybrid recommendation approach that combines social behaviors, the genres of movies and existing collaborative filtering algorithms to perform movie recommendation. The experiments with MovieLens dataset show the advantage of our proposed method comparing to the benchmark method in terms of recommendation accuracy.

Chen Yang, Xiaohong Chen, Lei Liu, Tingting Liu, Shuang Geng

Multi-feature Collaborative Filtering Recommendation for Sparse Dataset

Collaborative filtering algorithms become losing its effectiveness on case that the dataset is sparse. When user ratings are scared, it’s difficult to find real similar users, which causes performance reduction of the algorithm. We here present a 3-dimension collaborative filtering framework which can use features of users and items for similarity computation to deal with the data sparsity problem. It uses feature and rating combinations instead of only ratings in collaborative filtering process and performs a more complete similarity computation. Specifically, we provide a weighted feature form and a Bayesian form in its implementation. The results demonstrate that our methods can obviously improve the performance of collaborative filtering when datasets are sparse.

Zengda Guan

A Collaborative Filtering Algorithm Based on Attribution Theory

The Collaborative filtering algorithm predicts the user’s preference for the project to complete a recommendation by analyzing the user preference data, and usually takes the user’s rating as the user preference data. However, there is a bias between user’s preference and user’s score of the real scene, so the user’s rating as user preference can lead to lower recommendation accuracy. For this problem, this paper proposes a user preference extraction method based on attribution theory, calculates user preferences by analyzing user rating behavior. Then, combining preference similarity and rate similarity, making up the bias between user rating and user preference in collaborative filtering algorithm. Experimental verification on universal Dataset Movies lens-1m results shows that the algorithm is preferable to the existing collaborative filtering algorithm.

Mao DeLei, Tang Yan, Liu Bing

Investigating the Effectiveness of Helpful Reviews and Reviewers in Hotel Industry

With the growth of e-commerce, online consumer reviews have become important attributes that influence purchasing decisions. Especially, hotel industry has strongly influenced by online reviews due that most tourists cannot experience all hotels personally and the service levels among hotels are very different. However, the flood of online consumer reviews has caused information overload, making it difficult for consumers to choose reliable reviews. Therefore, helpful remarks of hotel review should potentially have strong influence on users. Previous research focused on how to predict the helpful scores of reviews, but it has not explored the influence of reviews marked with helpfulness. The aim of this study is to investigate whether the helpful reviews and reviewers who contribute many reviews really have effects on the marks hotel received. With analysis of reviews contributed in Tripadvisor.com for three hundred hotels scattered in ten cities of U.S., this study found both reviewer contribution, and helpful review has a positive effect on marks of hotels. Moreover, the research also discovered that the helpfulness of reviews is negatively relates to the ratings. Also, the research found that the standard deviation of review mark is positively relates to hotel ranks.

Yen Tzu Chao, Ping Yu Hsu, Ming Shien Cheng, Hong Tsuen Lei, Shih Hsiang Huang, Yen-Huei Ko, Grandys Frieska Prassida, Chen Wan Huang

Mapping the Landscapes, Hotspots and Trends of the Social Network Analysis Research from 1975 to 2017

A Bibliometric analysis was applied in this paper to quantitatively evaluate the social network analysis research from 1975 to 2017 based on 7311 bibliographic records collected from the Science Citation Index (SCI) database. Firstly, a comprehensive analysis was conducted to reveal the current landscapes such as scientific outputs, international collaboration, subject categories, and research performances by individuals, we then use innovative methods such as Burst Detection, Referenced Publication Years Spectroscopy and Keyword Semantic Clustering to provide a dynamic view of the evolution of social network analysis research hotpots and trends from various perspectives. Results shows that social network analysis research has developed rapidly in the past four decades and is in the growth period with a maturity of 50.00%, the total of 7311 articles cover 120 countries (regions) and the top five most productive countries are USA, England, China, Canada and Germany. Among the 1181 major journal related to social network analysis, University of Illinois, University of Sydney and Carnegie Mellon University ranked as the top three. In addition burst keywords such as Knowledge Management, Centrality, Modularity, Community, Link Prediction, Learning Analytics and Big Data demonstrate the trends of this field. The result provides a dynamic view of the evolution of Social Network Analysis research landscapes, hotspots and trends from various perspectives which may serve as a potential guide for future research.

Li Zeng, Zili Li, Zhao Zhao, Meixin Mao

Predication

Frontmatter

A Deep Prediction Architecture for Traffic Flow with Precipitation Information

Traffic flow prediction is an important building block to enabling intelligent transportation systems in a smart city. An accurate prediction model can help the governors make reliable traffic control strategies. In this paper, we propose a deep traffic flow prediction architecture P-DBL, which takes advantage of a deep bi-directional long short-term memory (DBL) model and precipitation information. The proposed model is able to capture the deep features of traffic flow and take full advantage of time-aware traffic flow data and additional precipitation data. We evaluate the prediction architecture on the dataset from Caltrans Performance Measurement System (PeMS) and the precipitation dataset from California Data Exchange Center (CDEC). The experiment results demonstrate that the proposed model for traffic flow prediction obtains high accuracy compared with other models.

Jingyuan Wang, Xiaofei Xu, Feishuang Wang, Chao Chen, Ke Ren

Tag Prediction in Social Annotation Systems Based on CNN and BiLSTM

Social annotation systems enable users to annotate large-scale texts with tags which provide a convenient way to discover, share and organize rich information. However, manually annotating massive texts is in general costly in manpower. Therefore, automatic annotation by tag prediction is of great help to improve the efficiency of semantic identification of social contents. In this paper, we propose a tag prediction model based on convolutional neural networks (CNN) and bi-directional long short term memory (BiLSTM) network, through which, tags of texts can be predicted efficiently and accurately. By Experiments on real-world datasets from a social Q&A community, the results show that the proposed CNN-BiLSTM model achieves state-of-the-art accuracy for tag prediction.

Baiwei Li, Qingchuan Wang, Xiaoru Wang, Wei Li

Classification

Frontmatter

A Classification Method for Micro-Blog Popularity Prediction: Considering the Semantic Information

Predicting the scale and quantity of reposting in micro-blog network have significances to the future network marketing, hot topic detection and public opinion monitor. This study proposed a novel two-stage method to predict the popularity of a micro-blog prior to its release. By focusing on the text content of the specific micro-blog as well as its source of publication (user’s attributes), a special classification method—Labeled Latent Dirichlet allocation (LLDA) was trained to predict the volume range of future reposts for a new message. To the authors’ knowledge, this paper is the first research to utilize this multi-label text classifier to investigate the influence of one micro-blog’s topic on its reposting scale. The experiment was conducted on a large scale dataset, and the results show that it’s possible to estimate ranges of popularity with an overall accuracy of 72.56%.

Lei Liu, Chen Yang, Tingting Liu, Xiaohong Chen, Sung-Shun Weng

VPSO-Based CCR-ELM for Imbalanced Classification

In class-specific cost regulation extreme learning machine (CCR-ELM) for the class imbalance problems, the key parameters, including the number of hidden nodes, the input weights, the hidden biases and the tradeoff factors are normally chosen randomly or preset by human. This made the algorithm responding slowly and generalization worse. Unsuitable quantity of hidden nodes might form some useless neuron nodes and make the network complex. So an improved CCR-ELM based on particle swarm optimization with variable length is present. Each particle consists of above key parameters and its length varies with the number of hidden nodes. The experimental results for nine imbalance dataset show that particle swarm optimization with variable length can find better parameters of CCR-ELM and corresponding CCR-ELM had better classification accuracy. In addition, the classification performance of the proposed classification algorithm is relatively stable under different imbalance ratios.

Yi-nan Guo, Pei Zhang, Ning Cui, JingJing Chen, Jian Cheng

An Ensemble Classifier Based on Three-Way Decisions for Social Touch Gesture Recognition

Touch is an important form of social interaction. In Human Robot Interaction (HRI), touch can provide additional information to other modalities, such as audio, visual. In this paper, an ensemble classifier based on three-way decisions is proposed to recognize touch gestures. Firstly, features are extracted from six perspectives and four classifiers are constructed on different scales with different preprocessing methods. Then an ensemble classifier is used to combine the four classifiers to classify touch gestures. Our method is tested on the public Corpus of Social Touch (CoST) dataset. The experiment results not only verify the validity of our method but also show a better performance of our ensemble classifier.

Gangqiang Zhang, Qun Liu, Yubin Shi, Hongying Meng

Engineering Character Recognition Algorithm and Application Based on BP Neural Network

Character recognition algorithm can directly affect the accuracy and speed of character recognition. This algorithm uses BP neural network to train samples, preserve neural network weights, and recognize photographed images. The software algorithm integrates image-processing and neural network modules. Image-processing modules include pre-treatment processes, such as, binaryzation, denoising, dilation, erosion, rotation and character segmentation and extraction of images collected by cameras. Neural network modules include network training, identification, display, saving, loading, and other modules, such as image preprocessing and recognition. A prototype of online engineering character recognition system has been developed. Test results indicate that the duration of a single picture is approximately 100 ms, and the detection time displayed by the interface includes the zooming time of display interface that is approximately 200 ms.

Chen Rong, Yu Luqian

Hand Gesture Recognition Based on Multi Feature Fusion

In view of the influence of complex and changeable gestures on recognition, a gesture recognition method based on multi feature phase fusion is proposed. Firstly, the skeleton feature and contour feature of the gesture area are extracted. Then the feature fusion method is used to obtain the fusion features of the gestures. Finally, support vector machine, decision tree, random forest and convolution neural network are used to recognize the skeleton feature, contour feature and fusion feature of gesture area respectively. The results show that under different data sets, gesture recognition based on multi feature fusion improves the recognition accuracy by 2% compared with single feature recognition algorithm, reaching 98.57%.

Hongling Yang, Shibin Xuan, Yuanbin Mo

Application of SVDD Single Categorical Data Description in Motor Fault Identification Based on Health Redundant Data

The system’s self-protection mechanism immediately stops the motor when motor system in the event of a malfunction, so it is difficult to collect the fault data when monitoring the motor status. Under the premise of only collecting motor’s health data, using SVDD algorithm to train health data and building non-health data sets based on practical experience in this paper. Based on BP neural network, a random self-adapting particle swarm optimization algorithm (RSAPSO) is used to substitute the original gradient descent method in BP network, training speed and accuracy of BP network training is improved. Three commonly used test functions were used to test the performance of the improved PSO, and the improved particle swarm optimization is compared with the standard particle swarm optimization, particle swarm optimization with compression factor and adaptive particle swarm optimization. In this paper, three asynchronous motor Y225S-4 output shaft vibration acceleration signal in healthy state as a case to test the effectiveness of the algorithm, results show that in the case of only health data, the new algorithm based on single classification has better performance and can effectively monitor the working state of the motor.

Jianjian Yang, Xiaolin Wang, Zhiwei Tang, Zirui Wang, Song Han, Yinan Guo, Miao Wu

Finding Patterns

Frontmatter

Impact of Purchasing Power on User Rating Behavior and Purchasing Decision

Recommender system have broad and powerful applications in e-commerce, news promotion and online education. As we all know, the user’s rating behavior is generally determined by subjective preferences and objective conditions. However, all the current studies are focused on subjective preferences, ignoring the role of the objective conditions of the user. The user purchasing power based on price is the key objective factor that affects the rating behavior and even purchasing decision. Users’ purchasing decisions are often affected by the purchasing power, and the current researches did not take into account the problem. Thus, in this paper, we consider the influence of user preferences and user purchasing power on rating behavior simultaneously. Then, we designed a reasonable top-N recommendation strategy based on the user’s rating and purchasing power. Experiments on Amazon product dataset show that our method has achieved better results in terms of accuracy, recall and coverage. With ever larger datasets, it is important to understand and harness the predictive purchasing power on the users’ rating behavior and purchasing decisions.

Yong Wang, Xiaofei Xu, Jun He, Chao Chen, Ke Ren

Investigating the Relationship Between the Emotion of Blogs and the Price of Index Futures

As the financial derivatives tradable market developed quickly in Taiwan, the trading volumes in futures grew quickly in recent years. At the same time, many people posted and shared opinion on social media. Many research in economics and behavioral finance have posited and confirmed that investor’s “mood” correlated with the performance of financial market. Several researches had devoted to study the relationship between the volatility of financial market and sentiments expressed in social media. On the other hand, even though emotion can describe the feeling of people more precisely than sentiment, to the best of our knowledge, only one research has tried to discover the relationship between futures performance and emotion fluctuation. The research tracked the evolvement of specific events. Instead of tracking long-term emotional fluctuation, this study strived to predict price change of derivatives with emotion expressed in social media in previous day. The result confirmed that there was a significant correlation between the intensity of emotion “fear” and the market decline. When the major emotions were “good” and “sad”, the strength of emotion was significantly correlated with the change of the market price.

Yen Hao Kao, Ping Yu Hsu, Ming Shien Cheng, Hong Tsuen Lei, Shih Hsiang Huang, Yen-Huei Ko, Chen Wan Huang

A Novel Model for Finding Critical Products with Transaction Logs

For the consumer market, finding valuable customers is the first priority and is assumed to assist companies in obtaining more profit. If we could discover critical products that are related with valuable customers, then it will lead to better marketing strategy to fulfill those essential customers. It will also assist companies in business development. This study selects real retail transaction data via the recency, frequency, and monetary (RFM) analysis and adopts the K-means algorithm to obtain results. Moreover, the Apriori algorithm with minimum support and skewness criteria is used to filter and find critical products. In this research, we found a novel methodology through setting the minimum support and skewness criteria and utilized the Apriori algorithm to identify 31 single critical products and 60 critical combinations (two products). This study assist companies in finding critical products and important customers, which is expected to provide an appropriate customer marketing strategy.

Ping Yu Hsu, Chen Wan Huang, Shih Hsiang Huang, Pei Chi Chen, Ming Shien Cheng

Using Discrete-Event-Simulation for Improving Operational Efficiency in Laboratories: A Case Study in Pharmaceutical Industry

Just-in-time delivery has become a key aspect of pharmaceutical industry when loyalizing customers and competing internationally. Additionally, prolonged lead times may lead to increased work-in-process inventory, penalties for non-compliance and cost overrun. The problem is more complex upon considering a wide variety of products as often noted in pharmaceutical companies. It is then relevant to design strategies focusing on improving the delivery performance. Therefore, this paper proposes the use of Discrete-event simulation (DES) to identify inefficiencies and define solutions for the delivery problem. First, input data were gathered and analyzed. Then, a DES model was developed and validated. Finally, potential improvement scenarios were simulated and analyzed regarding productivity rate and proportion of tardy jobs. A case study in a pharmaceutical laboratory is presented to validate the proposed methodology. The results evidenced that, by implementing the best scenario, the productivity may be augmented by 44.83% which would generate zero tardy jobs.

Alexander Troncoso-Palacio, Dionicio Neira-Rodado, Miguel Ortíz-Barrios, Genett Jiménez-Delgado, Hugo Hernández-Palma

Architecture of an Object-Oriented Modeling Framework for Human Occupation

The limitations of the actual theoretical structure of occupational science are discussed emphasizing on its implications when dealing with the stability and sustainability of social systems. By using a literature review focused on the time evolution and disciplinary distribution of the scientific production about human occupation, it is verified the insufficient production leading to the development of models that facilitate quantitative reasoning to support decision making. As an alternative, the architecture of an object-oriented framework is proposed. The framework is presented by using an UML (Unified Modeling Language) class diagram of a generic occupational system, including the class model of each system’s component: attributes and behaviors. Finally, guidelines are given for the use of the models produced with the framework in simulating diverse occupation systems scenarios.

Manuel-Ignacio Balaguera, María-Cristina Vargas, Jenny-Paola Lis-Gutierrez, Amelec Viloria, Luz Elena Malagón

A Building Energy Saving Software System Based on Configuration

A design method of building energy saving software system based on configuration is proposed, it can meet the needs of large-scale building energy efficiency through this method. The software system utilizes the configuration design concept to realize the process monitoring, analysis and evaluation functions for large-scale building energy consumption. It can find abnormal energy consumption equipment within the building, and reduce the peak power consumption to achieve the purpose of building energy efficiency. This paper first introduces the process control software development based on the idea of configuration, the overall framework of the building energy-saving configuration software system and the design process of each module is described in detailed. The software design practice validates the availability and good scalability of process control software based on configuration ideas. This system meets the needs of large building energy consumption monitoring and building energy efficiency.

Jinlong Chen, Qinghao Zeng, Hang Pan, Xianjun Chen, Rui Zhang

Measures of Concentration and Stability: Two Pedagogical Tools for Industrial Organization Courses

This document describes two pedagogical tools developed for teaching applied microeconomics, specifically the issues related to concentration, dominance, stability and asymmetry of firms. The tools make a compilation of several concentration and stability indexes used in the literature since 1945. Among the benefits of the applications are the ease and agility to perform comparative analyzes of intersectorial and/or intertemporal type in a simple and agile way; and the use of unconventional concentration and stability measures.

Jenny-Paola Lis-Gutiérrez, Mercedes Gaitán-Angulo, Linda Carolina Henao, Amelec Viloria, Doris Aguilera-Hernández, Rafael Portillo-Medina

Image Enhancement

Frontmatter

The Analysis of Image Enhancement for Target Detection

In the process of automatic detection and recognition based on image, the quality of the detected images affects the target detection and recognition results. To solve the problem of low contrast and high signal-to-noise ratio of the target image in the target detection process, this paper introduces two types of image detail enhancement algorithms which are widely used in recent years, including brightness contrast image enhancement algorithm and HSV color space based enhancement algorithm, and its impact on the target detection. Experiments show that the image detail enhancement can improve the overall and local contrast of the image, highlight the details of the image, and the enhanced image can effectively improve the number and accuracy of the target detection.

Rui Zhang, Yongjun Jia, Lihui Shi, Hang Pan, Jinlong Chen, Xianjun Chen

Image Filtering Enhancement

With the development of science and technology, mankind has entered the information age. Image has become the main source of human access to information. However, in the actual process of image signal transmission, the loss and damage of data packet are inevitable due to the physical defects of the channel, which lead to a serious decline in the quality of the video stream. So it is necessary and even urgent now to do some research work on image enhancement technology. In this paper, the image enhancement algorithms that are commonly used, such as bilateral filtering algorithm, homomorphic filtering algorithm, are analyzed in image processing. In the design of the image enhancement, the best modeling and design schemes are chosen according to the comparison. The experimental results demonstrate that the bilateral filtering algorithm can effectively maintain the details of the image edges and make the image more smooth; the homomorphic filtering algorithm can effectively adjust the image gray range, so that the image details on the image area can be increased, and the algorithm can handle the image with inhomogeneous intensity. This work will lay a good foundation of further research.

Zhen Guo, Hang Pan, Jinlong Chen, Xianjun Chen

Random Forest Based Gesture Segmentation from Depth Image

Gesture image segmentation is a challenge task due to the high degree of freedom of human gestures, large differences in shape and high flexibility, traditional pattern recognition and image processing methods are not effective in gesture detection. The traditional image segmentation based on the detection of skin color and the image of the depth image are limited by the effects of ambient light, skin color difference and image depth variation, resulting in unsatisfactory results. Therefore, we propose a hand gesture depth image segmentation method based on random forest. The method learns the gesture image feature representation of the depth image by supervising learning. Experiments show that the proposed method segments the gesture s’ pixels from the backgrounds area of the depth image. The proposed method potential has widely usages in gesture tracking, gesture recognition and human computer interaction.

Renjun Tang, Hang Pan, Xianjun Chen, Jinlong Chen

Deep Learning

Frontmatter

DL-GSA: A Deep Learning Metaheuristic Approach to Missing Data Imputation

Incomplete data has emerged as a prominent problem in the fields of machine learning, big data and various other academic studies. Due to the surge in deep learning techniques for problem-solving, in this paper, authors have proposed a deep learning-metaheuristic approach to combat the problem of imputing missing data. The proposed approach (DL-GSA) makes use of the nature inspired metaheuristic, Gravitational search algorithm, in combination with a deep-autoencoder and performs better than existing methods in terms of both accuracy and time. Owing to these improvements, DL-GSA has wider applications in both time and accuracy sensitive areas like imputation of scientific and research datasets, data analysis, machine learning and big data.

Ayush Garg, Deepika Naryani, Garvit Aggarwal, Swati Aggarwal

Research on Question-Answering System Based on Deep Learning

With the continuous development of the network, Question-Answering system has become a way for people to get information quickly. The QA task aims to provide precise and quick answers to user questions from a collection of documents or a database. In this paper, we introduce an attention based deep learning model to match the question and answer sentence. The proposed model employs a bidirectional long-short term memory(BLSTM) to solve the problem of lack features. And we also use the attention mechanism which allows the question to focus on a certain part of the candidate answer. Finally, we evaluate our model and the results show that our approach outperforms the method of feature construction based on machine learning. And the attention mechanism improves the matching accuracy.

Bo Song, Yue Zhuo, Xiaomei Li

A Deep Learning Model for Predicting Movie Box Office Based on Deep Belief Network

For the limitation that Chinese movie box office forecasting accuracy is not high in the long-term prediction research, based on the research of the Chinese movie market, this paper proposes a long-term prediction model for movie box office based on the deep belief network. The new model improved the movie box office influence model of Barry, screened out the effective box office impact factor, normalized the quantitative factor and formed a measurement system which is suitable for the Chinese movie market. Based on this measurement system, the characteristics of the data set in the original space are transferred to the space with semantic features and a hierarchical feature representation by deep learning, thus the accuracy of box office prediction was improved. Experimental evaluation results show that, in view of the 439 movie data, the DBN prediction model of movie box office has better prediction performance, and has good application value in the field of film box office.

Wei Wang, Jiapeng Xiu, Zhengqiu Yang, Chen Liu

A Deep-Layer Feature Selection Method Based on Deep Neural Networks

Inspired by the sparse mechanism of the biological nervous system, we propose a novel feature selection algorithm: features back-selection (FBS) method, which is based on the deep learning architecture. Compared with the existing feature selection method, this method is no longer a shallow layer approach, since it is from the global perspective, which traces back step by step to the original key feature sites of the raw data by the abstract features learned from the top of the deep neural networks. For MNIST data, the FBS method has quite well performance on searching for the original important pixels of the digit data. It shows that the FBS method not only can determine the relevant features for learning task with keeping a quite high prediction accuracy, but also can reduce the space of data storage as well as the computational complexity.

Chen Qiao, Ke-Feng Sun, Bin Li

Video Vehicle Detection and Recognition Based on MapReduce and Convolutional Neural Network

With the rapid growth of traffic video data, it is necessary to improve the computing power and accuracy of image processing. In this paper, a video vehicle detection and recognition system based on MapReduce and convolutional neural network is proposed to reduce the time-consume and improve the recognition accuracy in video analysis. First, a fast and reliable deep learning algorithm based on YOLOv2 is used to detect vehicle in real-time. And then the license plate recognition algorithm based on improved convolutional neural network is presented to recognize the license plate image extracted from the detected vehicle region. Finally, the Hadoop Video Processing Interface (HVPI) and MapReduce framework are combined to apply the video vehicle detection and recognition algorithms for parallel processing. Experimental results are presented to verify that the proposed scheme has advantages of high detection rate and high recognition accuracy, and strong ability of data processing in large-scale video data.

Mingsong Chen, Weiguang Wang, Shi Dong, Xinling Zhou

A Uniform Approach for the Comparison of Opposition-Based Learning

Although remarkable progress has been made in the application of opposition-based learning in recent years, the complete theoretical comparison is seldom reported. In this paper, an evaluation function of opposition strategy is defined and then a uniform evaluation approach to compute the mean minimum Euclidean distance to the optimal solution is proposed for one dimensional case. Thus different opposition strategies can be compared easily by means of the mathematical expectation of these evaluation functions. Theoretical analysis and simulation experiments can support each other, and also show the effectiveness of this method for sampling problems.

Qingzheng Xu, Heng Yang, Na Wang, Rong Fei, Guohua Wu

Backmatter

Weitere Informationen

Premium Partner

    Bildnachweise