Computational Linguistics and Information Processing
Over the past two decades, cities across the country have experienced a tremendous growth in cycling. As cities expand and improve their bicycle networks, local governments and bicycle associations are looking into ways of making cycling in urban areas safer. However, one of the main obstacles in decreasing the number of bicycle crashes is the lack of information regarding cycling safety at the street level. Historically, Bicycle Level of Service (BLOS) models have been used to measure street safety. Unfortunately, these models require extensive information about each particular roadway section, which often times is not available. This project provides innovative tools to automatically estimate street safety levels from crowdsourced citizens' complaints as well as to shed some light into the traffic-related reasons behind such safety values. Ultimately, the outcomes of this project will contribute to the overall vision for Smart and Connected Communities (S&CC) by helping to reduce the number of crashes and human fatalities in the city using large streams of data collected from connected citizens. The project has strong support from multiple local institutions including Bike Share and local transportation departments.
From a technical perspective, the main innovation will be the ability to automatically compute cycling safety measures using information extracted from citizen-generated complaints at very fine- grained spatio-temporal scales. For that purpose, the project will use data mining and machine- learning techniques to extract relevant quantitative and textual features from the crowdsourced data. The expected outcomes of this project will be: (a) accurate and interpretable models to estimate street safety levels from user-generated data; (b) a set of easy-to-interpret, actionable items for local Departments of Transportation to improve cycling experiences and general safety; and (c) a dataset with user-generated complaints, cycling videos and safety levels per road segments to share with other researchers so as to advance the state of the art in data-driven cycling safety.