مقاله انگلیسی رایگان در مورد رویکرد محاسباتی نرم برای خلاصه سازی کلان داده ها – الزویر 2018

The added value of a dataset lies in the knowledge a domain expert can extract from it. Considering the continuously increasing volume and velocity of these datasets, efficient tools have to be defined to generate meaningful, condensed and human-interpretable representations of big datasets. In the proposed approach, soft computing techniques are used to define an interface between the numerical and categorical space of data definition and the linguistic space of human reasoning. Based on the expert’s own vocabulary about the data, a personal summary composed of linguistic terms is efficiently generated and graphically displayed as a term cloud offering a synthetic view of the data properties. Using dedicated indexing strategies linking data and their subjective linguistic rewritings, exploration functionalities are provided on top of the summary to let the user browse the data. Experimentations confirm that the space change operates in linear time wrt. the size of the dataset making the approach tractable on large scale data. © 2018 Elsevier B.V. All rights reserved.

Introduction

Data analysis is a crucial task at the center of many professional activities and now constitutes a support for decision making, communicating and reporting. Considering the continuously increasing volume and velocity of these datasets, domain experts (as insurers, data journalists, communication managers, decision makers, etc.), who are not, most of the time, data or computer scientists, need efficient tools that help them turn data into useful knowledge. This explains the recent growing interest for so-called Agile Business Intelligence (ABI) systems that reconsider classical data integration processes to favor pragmatic approaches that make domain experts self-reliant in the analysis of raw data. A dataset generally consists of a large collection of items described by numerical and categorical attributes. A way to assist experts in their fastidious task of data-to-knowledge translation is to define efficient strategies that generate meaningful, condensed and human-interpretable representations of the data. To be very useful, such representations should give an insight into the data properties and make it easy for the domain expert to identify the most representative properties of the dataset. In this sense, when a dataset is so large that it cannot be easily perused and analyzed by the user, data summarization is of a particular interest to obtain a big picture of the data distribution on the different dimensions. Such a summary should also offer exploration functionalities to let the expert interactively browse the dataset from its summary and discover interesting properties possessed by different data subsets.