DEEPEYE: An Automatic Big Data Visualization Framework

作     者：Xuedi Qin Yuyu Luo Nan Tang Guoliang Li 

作者机构：Department of Computer Science Tsinghua University Qatar Computing Research InstituteHBKU 

出 版 物：《Big Data Mining and Analytics》 (大数据挖掘与分析（英文）)

年 卷 期：2018年第1卷第1期

页      面：75-82页

核心收录：

学科分类：0810[工学-信息与通信工程] 1205[管理学-图书情报与档案管理] 12[管理学] 1201[管理学-管理科学与工程(可授管理学、工学学位)] 0839[工学-网络空间安全] 0811[工学-控制科学与工程] 0812[工学-计算机科学与技术（可授工学、理学学位）] 

基　　金：supported by the National Key Basic Research and Development(973)Program of China(No.2015CB358700) the National Natural Science Foundation of China(Nos.61373024,61632016,61422205,and 61472198) 

主　　题：big data automatic data visualization visualization verification visualization ranking visualization search space 

摘      要：Data visualization transforms data into images to aid the understanding of data; therefore, it is an invaluable tool for explaining the significance of data to visually inclined people. Given a(big) dataset, the essential task of visualization is to visualize the data to tell compelling stories by selecting, filtering, and transforming the data, and picking the right visualization type such as bar charts or line charts. Our ultimate goal is to automate this task that currently requires heavy user intervention in the existing visualization systems. An evolutionized system in the field faces the following three main challenges:(1) Visualization verification: to determine whether a visualization for a given dataset is interesting, from the viewpoint of human understanding;(2) Visualization search space: a boring dataset may become interesting after an arbitrary combination of operations such as selections,joins, and aggregations, among others;(3) On-time responses: do not deplete the user s patience. In this paper,we present the DEEPEYE system to address these challenges. This system solves the first challenge by training a binary classifier to decide whether a particular visualization is good for a given dataset, and by using a supervised learning to rank model to rank the above good visualizations. It also considers popular visualization operations, such as grouping and binning, which can manipulate the data, and this will determine the search space. Our proposed system tackles the third challenge by incorporating database optimization techniques for sharing computations and pruning.