Table of contents

Preface

The 2017 4rd Big Data Conference (BDC 2017, formerly International Conference on Big Data and Its Applications) was successfully held on September 15, 2017 in Moscow, Russia. The conference provided a leading forum for disseminating the latest results in Big Data Research, Development, and Applications from business, technological and scientific points of view. For the scientific track of the conference solicit high-quality original research papers (and significant work-in-progress papers) in any aspect of Big Data with emphasis on 5Vs (Volume, Velocity, Variety, Value and Veracity), including the Big Data challenges in scientific and engineering, social, sensor/IoT/IoE, and multimedia (audio, video, image, etc.) big data systems and applications. There were 3 major sections of the scientific track of the conference:

- Fundamentals of Data Science

- Big Data Infrastructure

- Big Data Security, Privacy and Trust

- Big Data Management

- Big Data Search and Mining

- Big Data Applications

This volume of Journal of Physics: Conference Series is a compilation of the accepted papers in BDC 2017 and represents contributions that were presented in the scientific track of the conference.

On behalf of the organization committee I would like to thank all of the conference sponsors, partners and volunteers who made conference possible.

Looking forward to meet you during BDC 2018.

On behalf of organizing and program committees of the BDC 2017

Igor Balk, BDC 2017 co-chair.

All papers published in this volume of Journal of Physics: Conference Series have been peer reviewed through processes administered by the proceedings Editors. Reviews were conducted by expert referees to the professional and scientific standards expected of a proceedings journal published by IOP Publishing.

An intelligent system for SAP ERP user support is proposed in this paper. It enables automatic replies on users' requests for support, saving time for problem analysis and resolution and improving responsiveness for end users. The system is based on an ensemble of machine learning algorithms of multiclass text classification, providing efficient question understanding, and a special framework for evidence retrieval, providing the best answer derivation.

Multidimensional space is a convenient means of representing large amounts of information. This fully applies to information on the elemental status of population groups. The novelty of the approach of this study is based on the fact that the totality of the weight parts of all elements of the periodic table together makes up the weight of a person. In a multidimensional space, organisms with the same weight have the same sum of coordinates and are located on one hyperplane. Since for any norms it is important to have a ratio between the quantities that reflect the content of the chemical elements – a ray becomes the standard, for each point of which the ratio between the coordinates is observed. Large amounts of data will adequately represent the proximity an organism to one or another class, that is, increasing the accuracy of diagnosing the elemental status. The algorithm for diagnosing, therefore, should include finding the corresponding hyperplane, the point of intersection with the ray and determining the proximity to this point.

The paper considers the problem of predicting the quality of metallurgical production in a general formulation, when a huge amount of historical data of technological processes and product quality can be partitioned into a set of discrete classes of efficiency. Assuming that in the production process the technological process route can be changed, the problem of choosing the optimal route becomes important. When constructing the decision tree that recognizes the class of efficiency of the technological process, a special criterion for optimality of the partitioning of the set of classes of efficiency into two classes is introduced corresponding to the left and right branches of the decision tree in the node under consideration. The partitioning obtained by the proposed approach is close to optimal and can form the basis for constructing a decision tree, with the help of which a route is chosen to continue processing.

The importance of ECG classification is very high now due to many current medical applications where this problem can be stated. Currently, there are many machine learning (ML) solutions which can be used for analyzing and classifying ECG data. However, the main disadvantages of these ML results is use of heuristic hand-crafted or engineered features with shallow feature learning architectures. The problem relies in the possibility not to find most appropriate features which will give high classification accuracy in this ECG problem. One of the proposing solution is to use deep learning architectures where first layers of convolutional neurons behave as feature extractors and in the end some fully-connected (FCN) layers are used for making final decision about ECG classes. In this work the deep learning architecture with 1D convolutional layers and FCN layers for ECG classification is presented and some classification results are showed.

In this paper we discuss use of machine learning methods such as self organizing maps, k-means and Ward's clustering to perform classification of universities based on their income. This classification will allow us to quantitate classification of universities as teaching, research, entrepreneur, etc. which is important tool for government, corporations and general public alike in setting expectation and selecting universities to achieve different goals.

This paper discuses impact of macroeconomics factor on the university budgeting. Modern developments in the area of data science and machine learning made it possible to utilise automated techniques to address several problems of humankind ranging from genetic engineering and particle physics to sociology and economics. This paper is the first step to create a robust toolkit which will help universities sustain macroeconomic challenges utilising modern predictive analytics techniques.

The paper is devoted to detailed investigation of users' behavior and level of expertise on online medical forums. Two popular forums were analyzed in terms of presence of experts who answer health related questions and participate in discussions. This study provides insight into the quality of medical information that one can get from the web resources, and also illustrates relationship between approved medical experts and popular authors of the considered forums. During experiments several machine learning and natural language processing methods were evaluated against to available web content to get further understanding of structure and distribution of information about medicine available online nowadays. As a result of this study the hypothesis of existing correlation between approved medical experts and popular authors has been rejected.

In our day Fourier analysis is commonly used in varies fields, that require information extraction form time series data. However, the Fourier technique has own pros and cons and we can found, that more contemporary approaches can give significant improvements in some cases. We take the example of astrophysical problem and try to resolve it with Fourier analysis in comparison with Fractal analysis. From our results one can find, that in cases of noise extraction Fractal techniques are more efficient and decrease the time of analysis of time series. The proposed approaches can be applied on other fields, to analyze the data sets with dramatically increased volume.

The article focuses on the description of a new block cipher. Due to the heightened interest in BigData the described cipher is used to encrypt big volumes of data in cloud storage services. The main advantages of the given cipher are the ease of implementation and the possibility of probabilistic encryption. This means that the text encryption will be different when the key is the same and the data is the same. So, the strength of the encryption is improved. Additionally, the ciphered message size can be hardly predicted.

An approach to data integration, aimed on the ontology-based intelligent search in Big Data, is considered in the case when information objects are represented in the form of relational databases (RDB), structurally marked by their schemes. The source of information for constructing an ontology and, later on, the organization of the search are texts in natural language, treated as semi-structured data. For the RDBs, these are comments on the names of tables and their attributes. Formal definition of RDBs integration model in terms of ontologies is given. Within framework of the model universal RDB representation ontology, oil production subject domain ontology and linguistic thesaurus of subject domain language are built. Technique of automatic SQL queries generation for subject domain specialists is proposed. On the base of it, information system for TATNEFT oil-producing company RDBs was implemented. Exploitation of the system showed good relevance with majority of queries.

A new method was developed with a cascading Bloom filter (CBF) for executing SQL queries in the Apache Spark parallel computing environment. It includes the representation of the original query in the form of several subqueries, the development of a connection graph and the transformation of subqueries, the definition of connections where it is necessary to use Bloom filters, the representation of the graph in terms of Spark. On the example of the query Q3 of the TPC-H test, full-scale experiments were carried out, which confirmed the effectiveness of the developed method.

The goal of our investigation is optimization of network work in virtual data center. The advantage of modern infrastructure virtualization lies in the possibility to use software-defined networks. However, the existing optimization of algorithmic solutions does not take into account specific features working with multiple classes of virtual network functions. The current paper describes models characterizing the basic structures of object of virtual data center. They including: a level distribution model of software-defined infrastructure virtual data center, a generalized model of a virtual network function, a neural network model of the identification of virtual network functions. We also developed an efficient algorithm for the optimization technology of containerization of virtual network functions in virtual data center. We propose an efficient algorithm for placing virtual network functions. In our investigation we also generalize the well renowned heuristic and deterministic algorithms of Karmakar-Karp.