Understanding the spatio-temporal characteristics of Twitter data with geotagged and non-geotagged content: two case studies with the topic of flu and Ted (movie)
Understanding the spatio-temporal characteristics of Twitter data with geotagged and non-geotagged content: two case studies with the topic of flu and Ted (movie)
复制标题
DOI：
10.1080/19475683.2017.1343257
复制
发表时间：
2017-01-01
期刊：
影响因子：
5
通讯作者：
Spitzberg, Brian
中科院分区：
文献类型：
--
作者：
Issa, Elias;Tsou, Ming-Hsiang;Spitzberg, Brian
带有地理标记的Twitter消息的动态特性为研究人员分析疾病爆发、环境变化和社会运动等事件的空间扩散提供了巨大的潜力。然而，与非地理标记数据相比，地理标记数据的比例非常小，而非地理标记的推文通常包含自动机器人产生的噪音，位置欺骗和人为错误。鉴于这些挑战，本研究旨在了解地理标记和非地理标记之间的Twitter扩散特性的差异。使用两个关键词“流感”和电影“泰德”从四个目标城市（圣地亚哥、洛杉矶、丹佛、纽约）收集推文，以代表美国的不同主题和地理区域。本研究提出了方法和分析框架，以过滤噪音，分析扩散过程的内部结构，并调查地理标记的推文的空间分布及其与土地利用类型的关联。结果表明，有地理标记的推文噪声更小，与事件的相关性更强，过滤后的非地理标记推文在趋势分析和内容分析中是有效的，Twitter中的主题选择与地理标记和非地理标记推文之间的相关性有关。此外，多个主题的地理标记的推文显示出显着的空间变化与土地利用分布和城市结构。
The dynamic characteristics of geotagged Twitter messages provide researchers with vast potential for analysing the spatial diffusion of events such as disease outbreaks, environmental changes and social movements. The percentage of geotagged data, however, is extremely small compared to non-geotagged data, whereas non-geotagged tweets often contain noises generated by automated robots, location spoofing and human-made mistakes. Given these challenges, this study aims to understand the difference in Twitter diffusion characteristics between geotagged and non-geotagged. Tweets were collected using two keywords 'flu' and movie Ted from four targeted cities (San Diego, Los Angeles, Denver, New York) to represent different topics and geographical areas in the United States. This study presents methodological and analytical frameworks to filter out noises, analyse the internal structure of the diffusion process, and investigate the spatial distribution of geotagged tweets and their associations with land-use types. Results indicate that geotagged tweets demonstrated less noise and stronger correlations to events, filtered non-geotagged tweets were effective in the trend and content analyses, and the topic choice in Twitter was associated with the correlation between geotagged and non-geotagged tweets. Further, geotagged tweets of multiple topics showed significant spatial variation related to the land-use distribution and the city structure.