Using Graphical Features To Improve Demographic Prediction From Smart Phone Data
Using Graphical Features To Improve Demographic Prediction From Smart Phone Data
复制标题
DOI：
10.1145/3068943.3068948
复制
发表时间：
2017-05
期刊：
影响因子：
--
通讯作者：
S. Akter;L. Holder
中科院分区：
文献类型：
--
作者：
S. Akter;L. Holder
人口统计信息，如性别、年龄、种族、教育水平、残疾、就业和社会经济地位，在社会科学、调查和营销领域都很重要。但由于用户不愿意参与，回复率低，很难从用户那里获得人口统计信息。通过智能手机传感器数据的自动人口统计预测，研究人员可以以非侵入性和成本效益的方式获得这些有价值的信息。我们通过使用基于图形特征的框架来处理人口预测问题，即性别、年龄组和工作类型的分类。该框架将从传感器网络中收集的信息表示为图形，提取有用和相关的图形特征，并预测人口统计信息。我们对诺基亚移动电话数据集的三个分类任务进行了评估：性别、年龄组和工作类型。我们的方法产生了与大多数最先进的方法相当的结果，同时具有普遍适用于传感器网络的额外优势，而无需使用复杂的和特定于应用的特征生成技术、背景知识和特殊技术来解决类不平衡问题。
Demographic information such as gender, age, ethnicity, level of education, disabilities, employment, and socio-economic status are important in the area of social science, survey and marketing. But it is difficult to obtain the demographic information from users due to reluctance of users to participate and low response rate. Through automated demographics prediction from smart phone sensor data, researchers can obtain this valuable information in a nonintrusive and cost-effective manner. We approach the problem of demographic prediction, namely, classification of gender, age group and job type, through the use of a graphical feature based framework. The framework represents information collected from sensor networks as graphs, extracts useful and relevant graphical features, and predicts demographic information. We evaluated our approach on the Nokia Mobile Phone dataset for the three classification tasks: gender, age-group and job-type. Our approach produced comparable results with most of the state of the art methods while having the additional advantage of general applicability to sensor networks without using sophisticated and application-specific feature generation techniques, background knowledge and special techniques to address class imbalance.