Association for Public Policy Analysis & Management

There is a growing interest in applying computational tools to the automatic discovery of social and economic behavior. For example, in many resource allocation problems, the ability to predict behavioral failures using real-time consumer data can allow for more efficient policy responses. In this paper, we use large-scale social data from a popular electric vehicle (EV) driver app to analyze the quality of charging services in the emerging EV charging infrastructure in the United States. We introduce a typology of EV charging experiences collected from a national dataset of 127,257 consumer reviews. We deploy various text classification algorithms, including convolutional neural networks (CNN), to automatically learn about potential market failures in sustainable transportation infrastructure from unstructured data. After classifying the electric vehicle user reviews into 9 main user topics and 34 subtopics, we find evidence of persistent quality issues in the existing infrastructure related to charging station functionality and lack of station availability, particularly in urban areas. Using this approach, we demonstrate the possibility to use machine learning tools to significantly reduce research evaluation time and cost from weeks of using human experts to classify unstructured data, to just minutes of computation. We discuss future directions for evaluating behavioral issues such as range anxiety, charge rage and congestion, which remain fundamental barriers to widespread EV adoption, and as a result, are of critical importance to achieving broader societal goals for air pollution and emissions reductions from the transportation sector.