A text mining approach to elicit public perception of bike-sharing systems

Bike-sharing has globally emerged as an alternative travel mode for trips that are longer to walk but shorter to drive. Previous studies have used either the actual ridership data or survey responses from users to understand the public perception about bike-sharing systems. Where the actual ridership data is hard to obtain, survey-based studies limit respondents' ability to express their views by relying on structured questionnaires (e.g., Likert scale). To this end, we contribute with the first application of a text network approach by analyzing the open-ended text responses from over 700 Seattle residents regarding their perceptions about the characteristics of docked and dockless bike-sharing systems. The text network approach enables the analysis of the open-ended text responses by creating a network based on the frequency and co-occurrence of keywords in a sentence. Our empirical analysis shows that Seattle residents appreciate dockless bike-sharing systems for their flexibility. However, they are unhappy about the blocked sidewalks due to parked bikes and less usage of helmets. Additionally, the text network's sparsity indicates that respondents have a variety of negative perceptions regarding docked bike-sharing systems, and therefore, improving these systems is challenging. The method also allows us to explore the heterogeneity in user groups' perceptions based on their bike-sharing experiences. Considering the consistency of our findings with those obtained using econometric approaches, we suggest that a hybrid approach can leverage the advantages of both econometric and text-based analysis. The approach would lead into reliable policy recommendations in the context of new services and technologies.