Transforming texts to maps: Geovisualizing topics in texts

Abstract

Unstructured textual data is one of the most dominant forms of communication. Especially after the adoption of Web 2.0, there has been a massive surge in the rate of generation of unstructured textual data. While a large amount of information is intuitively better for proper decision-making, it also means that it becomes virtually impossible to manually process, discover and extract useful information from textual data. Several supervised and unsupervised techniques in text mining have been developed to classify, cluster and extract information from texts. While text data mining provides insight to the contents of the texts, these techniques do not provide insights to the location component of the texts. In simple terms, text data mining addresses “What is the text about?” but fails to answer the “Where is the text about?” Since textual data have a large amount of geographic content (estimates of about 80%), it can be safely reasoned that answering “Where is the text about?” adds significant insights about the texts. In this study, a collection of news articles from the year 2017 were analyzed using topic modelling, an unsupervised text mining technique. Topics were discovered from the text collections using Latent Dirichlet Allocation method, a popular topic modelling technique. Topics are probability distribution of words which correspond to one of the concepts covered in the text. Spatial locations were extracted from text documents by geoparsing them. Topics were geovisualized as interactive maps according to the probability of each spatial location word which contributed to the corresponding topic. This is analogous to thematic mapping in Geographical Information System. Coordinates obtained from geoparsed words provide basis for georeferencing the topics while the probability of such location words corresponding to the particular topics provide the attribute value for thematic mapping. An interactive geovisualization of Choropleth maps at the level of country was constructed using the Leaflet visualization library. A comparative analysis between the maps and corresponding topics was made to see if the maps provided spatial context to the topics