Convolutional Neural Networks on Time Series for Smartphone Application Activations Using Wavelet Transform
Convolutional Neural Networks on Time Series for Smartphone Application Activations Using Wavelet Transform
复制标题
DOI：
10.1109/iiai-aai.2019.00112
复制
发表时间：
2019-07
期刊：
影响因子：
--
通讯作者：
T. Furuya;Hiromi Kondo;Fumiyo N. Kondo
中科院分区：
文献类型：
--
作者：
T. Furuya;Hiromi Kondo;Fumiyo N. Kondo
本研究的重点是智能手机应用程序激活（SAA）分析的时间序列。这些输入是从智能手机上的单个来源面板自动收集的三类时间序列的大数据。消费者激活输出的类型是预先分类的。本研究提取有效的功能，以确定类别的个人网站访问活动。这是数字营销时代的一项重要但具有挑战性的任务。大多数现有的研究依赖于传统的简单的统计分析，时间序列分析，或浅层神经网络，不能识别功能，以准确地分类多个类别的网站访问活动。本研究通过使用深度卷积神经网络（CNN）和贝叶斯优化，从SAA问题的原始输入中提出了一种自动特征学习模型。数据是未经转换或转换的两个层次的小波变换，为可能的应用在数字营销。
This research focuses on time series of smartphone application activation (SAA) analyses. The inputs are big data of three categories of time series automatically collected from a single source panel of individuals from smartphones. The types of consumer activation outputs are precategorized. This study extracts effective features to identify categories of individual website access activities. This is an important but challenging task in the age of digital marketing. Most existing studies rely on conventional simple statistical analyses, time-series analyses, or shallow neural networks that cannot identify features to accurately classify plural categories of website access activities. This study proposes an automatic feature learning model from the raw inputs for the SAA problem by using a deep convolutional neural network (CNN) and Bayes optimization. Data are non-converted or converted with two levels by wavelet transformation for possible application in digital marketing.