Delineating Urban Functional Areas with Sina Weibo Check-in Data: A Matching Time Series Distance Based LST-SVM Multi-classifier Method

Tianyu Xia ； Kai Yang ； Wanggen Wan ； Xiang Feng ； Xiaoqing Yu

《電腦學刊》 30卷6期 (2019/12) Pp. 181-194

https://doi.org/10.3966/199115992019123006014

摘要

Delineating the distribution of different urban functional areas is a hot topic in urban studies. The accuracy of describing urban functional areas is particularly important. This paper aims to improve the estimated accuracy of delineating urban functional areas according to the assumption that social media activities in these buildings with similar functionality not only have similar spatiotemporal patterns but also are strictly correlated to the temporal information. We propose a novel Matching Time Series (MTS) method to calculate the distance of the time-series data and use this method to modify the least squares twin support vector machine (LST-SVM) Multi-classifier Method for classifying the building objects with similar functionality as a functional area. According to the time series dataset built based on Sina Weibo check-in data, we compare with the dynamic time warping (DTW) distance based k-medoids method from different aspects. The results show that the accuracy is improved from 29.96% to 82.68%, which verifies the superiority of our proposed approach in improving the estimated accuracy of delineating urban functional areas. By separating time series dataset into weekdays and weekends, we also obtain relatively high classification accuracy respectively, and it contributes to analyze the distribution of urban functional buildings more clearly.

關鍵字

Delineating urban functional areas ； Sina Weibo check-in data ； matching time series ； Least squares twin support vector machine