Deep Learning in Urban Research; Exploring Local Issues By Mapping Semantic Themes and Sentiment

Friday, 20 July 2018: 08:42
Oral Presentation
Robin LYBECK, Åbo Akademi University, Finland
Deep learning in urban research; exploring local issues by mapping semantic themes and sentiment

In line with the development in other sciences, the study of large, varied and continuously growing data (i.e. Big data) has resulted in an increased adaption of data driven methods in social science research. Big data analysis has become increasingly topical in the study of spatial issues in the urban context. Municipalities around the world are opening up location based public data for researchers to utilize, and the software tools for analyzing large quantities of data becomes increasingly advanced and attainable. However, as new computational methods are constantly developed, the evaluation of their potential in research lags behind. Deep machine learning based on neural networks is quickly becoming mainstream in the analysis of everything from images to audio. The method has also proven successful in semantic analysis and predicting contextual themes from large corpora of text data. Combined with location based data the analysis can leap from semantic space to spatial space. In this case study of 34 000 citizen feedback messages in Turku, deep machine learning is combined with location based data analysis. Preliminary results of the case study show the prevalence of semantics relating safety and danger in citizen feedback relating to the urban environment. This article further addresses the problems relating to the use of deep machine learning for semantic text analysis and issues relating to the study spatially bound phenomenon in this way. The methodological concerns relating to this type of explorative data-driven analysis are highly topical in contemporary data-rich urban environment.