Premium
Novel data mining paradigms based on soft computing and machine learning in the current and upcoming information society revolution
Author(s) -
Choi Chang,
Pop Florin,
Huang Jun
Publication year - 2020
Publication title -
concurrency and computation: practice and experience
Language(s) - English
Resource type - Journals
SCImago Journal Rank - 0.309
H-Index - 67
eISSN - 1532-0634
pISSN - 1532-0626
DOI - 10.1002/cpe.5937
Subject(s) - computer science , library science , world wide web , artificial intelligence
The possibility of having the Internet accessibility everywhere and every time and the progressive reduction of the dimensions of hardware equipment, in addition to the advent of novel ICT paradigms realizing the Internet of Things and Smart City visions, made available the design and realization of an innovative type of applications where data play a key role, promising to achieve a non-negligible impact on the society and business processes. The right management, control, and distribution of data are extremely critical process in our society, so that it is always referred as information society. This is boosted by the proliferation of the social networks within our lives, and the upcoming of the Internet of Things within our homes and factories. Having such a vast volume of data to be processed in order to take decisions and/or infer knowledge poses a number of challenges. The issue of properly storing such data leads to the advent of different approaches than the traditional relational databases, such as the NoSQL products. The issue of supporting the scalability of the overall communication infrastructure and to balance the load of the consequent processing cause the shift from proprietary and centralized computing solutions to more distributed and multi-tenant approaches, such as the cloud computing and its variants. However, the processing is still centralized in the core of the infrastructures, and we only see a change on how such a core is concretely realized. Decentralizing the processing by distributing in along all the layers of the infrastructure is extremely needed, so as to handle the data scale and also the demand for timely responses. To let this to happen, novel processing approaches and schemes are emerging. We are witnessing an interplay among soft computing and machine learning for big data mining and processing.1 A non-marginal aspect is also on how to extract what is hidden in such a large amount of information, which is mostly non-structured or suitable for all the possible applications using them. It is extremely important to learn in an unsupervised manner from these data so as to extract the knowledge that they contain. A lot of historical data is currently available and storing all of them will surely reach the storage limits of the system. It is important to infer knowledge and keep only it instead of all the data. Last, policy makers need supports when taking decisions and such data can be of pivotal importance for them. In smart cities and factories certain decisions may be needed to be taken in an automatic manner so as to fast react to certain sudden and potentially critical events. It is extremely crucial understand how artificial intelligence can be used in such contexts and their possible limits. The aim of this special issue was to bring together experiences between soft computing and machine learning to the key issue of data mining to promote a convergence and cross-fertilization among them. To this Special Issue, 14 articles, as they present the most interesting research studies within the subject matter of this special issue. The paper “Personalized content recommendation scheme based on trust in online social networks” by Jaesoo Yoo et al2 presents a solution for content recommendation leveraging on trust estimations obtained from data extracted by online social network services. The behavior of a given person over the social network and the relationship with other users are collected and used to infer his/her trustworthiness by aggregating its experience and reputations. The proposed approach has been implemented and assessed, showing a reduced error than the main related works within the current literature. The paper “Deep Learning for EEG Data Analytics: A Survey” by Jason J. Jung et al3 analyses the current literature semi-supervised learning using EEG data analytics and points our pros and cons of these solutions to analyze EEG data. Such a study presents what are the main applications of EEG data analysis and provides concrete examples. The study ends by highlighting the future possible research directions in the context of analyzing EEG data, mainly by leveraging the recent findings deep learning, and possible additional applications. The paper “Extraction of abstracted sensory data to reduce the execution time of context-aware services in wearable computing environments” by Jongsun Choi et al4 presents a reducing method for the execution time of context-aware services. The proposed method gives a better performance without searching every single keyword element among the triplet sets. Also, they suggested method to the context-aware workflow middleware and demonstrated that the proposed method improves the processing time by at least 30% compared with the linear search by enhancing the comparison module in the middleware. The paper “Opcode sequence analysis of Android malware by a convolutional neural network” by Wenbo Shi et al5 discusses a new malware detection system for mobile devices running the Android Operative System. The proposed scheme uses an optimized deep convolutional neural network to infer knowledge from opcode sequences. The learning solution has been used in a supervised manner by training it multiple times using as input the raw opcode sequences obtained from a selected decompiled known Android malware. This allows to have a proper learning of the