The last decades have seen an enormous increase of digital text data from Internet sources such as social media platforms, blogs, web forums, web news, and so on. These data contain rich information about people’s personal perceptions, emotions, and opinions, as well as their activities and relationships. They are used in a large variety of fields, for instance, to assess human perception of their environment or gain situational awareness in disaster response (Z. Wang et al., 2019; Zhu et al., 2022). Some of these data also contain information on the geolocation of the messages, either in an explicit form (e.g., by using geotags) or in an implicit form (i.e., by mentioning locations in the texts), so that these messages can be located via geoparsing methods (Middleton et al., 2018), which in turn means that the texts can be analyzed spatially e.g., by means of geostatistics or point-pattern analyses (Cressie, 2015), or – if timestamps are available – by means of mobility analyses (Gonzalez et al., 2008). With using geographic location as a link, the content of the messages can be related to a wealth of geodata from other sources, such as volunteered geographic information (Goodchild, 2007), the internet of things (Kamilaris & Ostermann, 2018), or remote sensing imagery (H. Wang et al., 2018; Taubenböck et al., 2018). Combining such heterogeneous datasets offers new insights into how physical space and socially constructed space, i.e. place, interact. We call this new approach 'Geolingual Studies' (GLS), which integrates methods from the areas of artificial intelligence and digital humanities with those from linguistics, especially sociolinguistics, corpus-linguistics and (critical) discourse analysis, and remote sensing to investigate the relationship between physical space and place.
In this talk, we will illustrate the opportunities and challenges this framework offers with the help of several case studies: Firstly, we show how social media data and remote sensing data can be combined and contrasted to assess digital inequality between new and old urban spaces. For 100 settlements across Africa, we used satellite imagery to map the expansion of each settlement over time. This enables the comparison of tweet density between older and younger parts of the settlement. Results confirm the existence of a digital disparity between older and newer settlement areas that we found to be related, in complex ways, to settlement structure and the geographic setting (Mast et al., in preparation). The other two case studies presented here feature an additional layer of analysis by integrating a linguistic analysis of the textual data. To show how different topics (e.g., sports or politics) and events discussed by urban citizens can be identified and located in space, we present an analysis of the intra-urban diversity of topics discussed in geotagged Tweets from New York City (Lemoine-Rodríguez et al., in preparation). We show how topics and events can be identified across languages and characterized regarding the expressed emotions and opinions using a combination of natural language processing and qualitative analysis. By these means, discourses can be assessed across time, languages, and neighborhoods of the city. Finally, topics and languages can also be analyzed by incorporating mobility information at user-level, as we show in an application in a humanitarian crisis setting, i.e. the war in Ukraine. We identify migrant flows and the main needs and interests of migrants across space and migration stage, based on geotagged Twitter data (Lemoine-Rodríguez et al., 2024). The results show that the topics discussed by migrants on social media shifted depending on their migration stage (i.e., before leaving, after leaving, or after returning to Ukraine), and that the used language varied depending on the topic.
Challenges lie in the comparatively small proportion of text content which can be reliably geolocated (Zhu et al., 2022), the representativeness of the userbase (Lemoine-Rodriguez et al., 2024), the protection of users’ (geo-)privacy (Kounadi & Resch, 2018), and questions of ethics in this domain (Kochupillai et al., 2022). Additionally, as the example of Twitter demonstrates, the stability of data sources cannot be taken for granted (Davidson et al., 2023). However, many of the challenges associated with social media data can be effectively mitigated through technical methods. Furthermore, the comparison of results derived from social media with other datasets (e.g., official statistics) allows to confirm the plausibility of the insights derived from such data (Lemoine-Rodriguez et al., 2024). The benefits of social media data for research are substantial. Social media represents a rich data source for research as well as for decision-makers e.g., in crisis response situations, providing first-hand information in real-time of various facets of human behavior, including needs, opinions, interests, sentiments, and in some cases, mobility (Hübl et al., 2017; Mast et al., 2023; Zhu et al., 2022). In this sense, the combination of insights derived from text data and traditional data sources has great potential to improve our understanding of society. Thus, the combination of geographic and linguistic approaches can help to assess social behavior in a more holistic manner than any of these disciplines alone. These insights can be useful for psychologists, urban planners, or sociologists, and contribute meaningfully to several research disciplines, including the Spatial Humanities.