The International Archives of the Photogrammetry, Remote Sensing and Spatial Information Sciences
Download
Publications Copernicus
Download
Citation
Articles | Volume XLVIII-4/W1-2022
Int. Arch. Photogramm. Remote Sens. Spatial Inf. Sci., XLVIII-4/W1-2022, 315–320, 2022
https://doi.org/10.5194/isprs-archives-XLVIII-4-W1-2022-315-2022
Int. Arch. Photogramm. Remote Sens. Spatial Inf. Sci., XLVIII-4/W1-2022, 315–320, 2022
https://doi.org/10.5194/isprs-archives-XLVIII-4-W1-2022-315-2022
 
06 Aug 2022
06 Aug 2022

MAPPING THE CHATTER: SPATIAL METAPHORS FOR DYNAMIC TOPIC MODELLING OF SOCIAL MEDIA

L. Morandini, A. R. Mohammad, and R. O. Sinnott L. Morandini et al.
  • School of Computing and Information Systems, The University of Melbourne, Parkville VIC 3010, Melbourne, Australia

Keywords: Dynamic Topic Modelling, Social Media, Spatialization, Twitter, Information Visualization, Model Validation

Abstract. Topic modelling is a branch of Natural Language Processing (NLP) that deals with the discovery of conversation topics in a given document corpus. In social media, this translates into aggregating social media posts, e.g. tweets, into topics of conversation and observing how these topics evolve over time (hence the “dynamic” adjective). Conveying the results of topic modelling can be challenging since the topics often do not lend themselves naturally to meaningful labelling. The volume of real world (global) social media also means that millions of topics can be ongoing at any given time and the relationships between them can involve hundreds of dimensions and relationships that continually emerge. The popularity of topics is itself subject to change over time and reflect the pulse of what is happening in society at large. In this paper, we propose a spatialization technique based on open-source software that reduces the intrinsic complexity of dynamic topic modelling results to familiar topographic objects, namely: ridges, valleys, and peaks. This offers new possibilities for understanding complex relationships that change over time whilst overcoming issues with traditional topic modelling visualisation approaches such as network graphs.