利用中文微博数据的地理情感特征挖掘  

Mining Emotional Geography Features Based on Chinese Weibo Data

作　　者：王海起[1] 陈冉[1] 魏世清 桂丽[1] 费涛[1] WANG Haiqi;CHEN Ran;WEI Shiqing;GUI Li;FEI Tao(School of Ocean and Space Information,China University of Petroleum(East China),Qingdao 266580,China;School of Remote Sensing and Information Engineering,Wuhan University,Wuhan 430079,China)

机构地区：[1]中国石油大学(华东)海洋与空间信息学院,山东青岛266580 [2]武汉大学遥感信息工程学院,湖北武汉430079

出　　处：《武汉大学学报（信息科学版）》2020年第5期699-708,共10页Geomatics and Information Science of Wuhan University

基　　金：国家自然科学基金(41471322,41874146)。

摘　　要：情感是一类存在于空间和时间中的地理知识,但其获取存在难度。位置社交网络为情感的度量提供了优质数据源,基于2016年9月19日-25日和2016年10月1日-7日(国庆假期)两周发布的345万条带有地理位置的新浪微博数据,通过清洗、分词以及基于词典的情感度量方法,计算了每条微博的情感倾向,通过情感聚合探测了城市微博情感的空间分布特征,并在格网尺度上分析了微博情感在时空域中的热/冷点模式及趋势。研究表明:微博整体情感倾向以积极为主,个体情感的空间分布具有随机性,城市情感的空间分布极其不均匀,表现为局部地区情感高值/低值聚集区与低-高值/高-低值异常区伴随出现的分片分布特征,且日常时期这种局部差异性更为显著;以时空立方体为格网单元,发现中国东西部地区呈现对立的情感模式,东部以情感热点模式为主,且两个时期表现为不同的时空热点类型,西部则以情感冷点模式突出,且相比于日常时期,国庆假期的冷点模式存在加强趋势。研究结果揭示了微博情感在地理时空域分布的多样性和差异性,为分析中国居民幸福感、指导社会经济发展规划等提供辅助支持。Emotion is a kind of geography knowledge existing in space and time, but its acquisition is difficult. Location social networks provide high-quality data sources for emotion measurement. Based on approximately 3.45 million Sina Weibo data with geo-locations, this paper tweeted between September 19-25, 2016 and October 1-7, 2016. Firstly, through text cleaning, Chinese word segmentation and sentiment analysis based on dictionaries and grammar rules, the sentiment tendency of each Weibo text was scored.Then, the emotion values were aggregated at city level and spatial distribution characteristics of city Weibo emotions were detected. Finally, hot-spot and cold-spot spatiotemporal patterns and trends of Weibo emotions were analyzed at hexagon grid level. The research shows that the overall Weibo emotion tendency in China is positive and spatial distribution of individual emotions is random. At city level, the spatial distributions of emotions is extremely uneven and presents sub-regional distribution pattern, which high-high or lowlow clusters co-occur with low-high or high-low outliers in local areas and these local differentiations are more significant in September. At hexagon grid level, emotion patterns are opposite between the eastern and western regions of China. The eastern regions mainly present emotional hot-spot patterns, and September and October have different types of hot-spot patterns. In the western regions, emotional cold-spot patterns are prominent, and compared with September, cold-spot patterns of the National Day holiday tend to strengthen. The research reveals the diversities and differences of space-time distribution of Weibo emotions, which has instructive significance for analyzing the happiness of Chinese residents and social and economic development planning.

关 键 词：地理情感 微博文本 情感计算 空间和时空分布特征 热点和冷点探测 

分 类 号：P208[天文地球—地图制图学与地理信息工程]