Transforming Large-Scale Participation Data Through Topic Modelling in Urban Design Processes

The advancements in digital tools and data collection methods ensure the continuing growth of textual data obtained through large-scale participation processes in urban contexts. In order to extract the thematic content of such underutilized textual datasets, topic modeling (TM) and content analysis have been deployed as promising AI-based Natural Language Processing (NLP) techniques. Yet, implementing such techniques has not been exploited in urban design domains due to the complexity of textual datasets and the lack of a systematic evaluation framework. In this paper, we addressed the challenges in the utilization of large textual data by using a real-world dataset collected via a digital participation platform in Madrid, Spain. Firstly, we identified prominent data structures and potential information embedded into the dataset by using a document-oriented NoSQL database. In this step, we systematically discussed data pre-processing steps to convert them into a series of structured data collections. Secondly, we evaluated three different TM algorithms, i.e. LDA, LSI, and HDP, according to a number of hyperparameters controlling the learning process. This step aimed to reveal the required number of topics to extract meaningful content through the algorithms. Lastly, we presented possible textual data visualization techniques to enable the use of textual information in digital participation processes. Consequently, this paper facilitates the use of large textual datasets by investigating data structures & processing, revealing the potentials of different TM algorithms, and eventually analyzing the results with the support of urban big data analytics and computational linguistic techniques for informed urban design processes.