地理与地理信息科学

针对现有方法从微博中识别时空信息精度较低且相对粗略的问题,该文提出基于双向长短期记忆网络和条件随机场(BiLSTM-CRF)与分类分层标注的微博中突发事件时空信息精细识别方法(MFISIE)。首先,设计一套适用于微博中突发事件的分类分层时空信息标注体系(CHSIAS),构建微博语料库;然后,结合BiLSTM-CRF构建微博时空信息识别模型,并利用117 567条标注的微博语料进行实验。结果表明:与人民日报语料标注体系相比,CHSIAS与CRF、BiLSTM、BiLSTM-CRF 3种方法结合进行时空信息识别,F值均较高,且可获得多层级、精细化的突发事件时空信息,其中,基于BiLSTM-CRF的MFISIE方法的F值(91.2%)最高。使用CHSIAS时,BiLSTM-CRF对时间信息的识别效果最好,其对兴趣点、建筑物和相对位置描述识别结果的F值比BiLSTM方法分别提升了8.8%、6.3%和12.3%,比CRF方法分别提升了7.1%、7.7%和8.9%。MFISIE可更精确地提取微博中突发事件的时空信息,为突发事件应急信息的快速感知与精准应用提供技术支撑。

Weibo has the merits of being free, massive and up to date.It has become a new data source for obtaining emergency information.Timely collecting the information related to emergencies from Weibo helps decision makers to make decisions immediately and distribute emergency resources reasonably.Therefore, it is of great significance to study the methods of accurately and effectively identifying emergency information from Weibo.Moreover the spatio-temporal information is the key information for describing emergencies.For the problems of low precision and rough spatio-temporal information identified from Weibo by the existing methods, this paper proposes the MFISIE method for identifying fine spatio-temporal information of emergencies from Weibo.First, this paper designs a set of classified-hierarchical spatio-temporal information annotation system(CHSIAS) that is suitable for emergency information from Weibo, and constructs the Weibo emergencies corpus.Then, for identifying fine spatio-temporal information from Weibo, it builds the Weibo′s spatio-temporal information identification model combining with the bidirectional long-short term memory and conditional random field(BiLSTM-CRF).Finally, it carries out the experiments by using 117 567 lines of labeled Weibo corpus.Compared with the corpus annotation system of People′s Daily, the experimental results demonstrate that CHSIAS can obtain a higher F-measure and obtain multi-levels and refined spatio-temporal information of emergencies when it is combined with CRF,BiLSTM and BiLSTM-CRF respectively.Moreover, the F-measure of MFISIE is the highest, reaching 91.2%.MFISIE has the best performance of recognizing the time information, and for the position information, the F-measure of identifying POI,building and relative position based on MFISIE are 8.8%,6.3% and 12.3% higher than the BiLSTM-based results respectively, also 7.1%,7.7% and 8.9% higher than the CRF-based results respectively.Obviously, the MFISIE can effectively and accurately extract the fine spatio-temporal information of emergencies from Weibo, and can provide better technical support for the rapid perception and precise application of urban emergency information.