A Text Mining Approach to Discover Real-Time Transit Events from Twitter

The accelerated growth of the number of inhabitants in the cities brings with it the increase in the number of means of transport, generating new conflicts related to traffic and mobility. This growth and lack of alternative public transportation create a scenario where traffic becomes a serious problem. Such is the case in Cuenca, a city located in Ecuador with a population growth of 15% in the last 7 years, and so has the number of cars. Moreover, transit information is only delivered by traditional media which is not always accurate or in real-time. It is imperative to create a system to discover real-time events to help the population to acquire precise information. With the arising of social networks such as Twitter, new opportunities to solve the transit problem at its origin. Twitter users interact with the social network every day and inform their fellow users of different topics such as transit. We take Twitter as a source of information to feed a real-time system which infers transit data from tweets. We create a predictive model with the use of pre-processing techniques for data cleaning, Support Vector Machines for predictive modeling, dictionaries and Levenshtein distance for location discovery, and finally, association analysis for data pattern finding. Our results show that our approach outperforms the existing works in the field. Furthermore, we have achieved accuracy values greater than 90% in classification subroutines and more than 70% in location discovery. Thus, we have settled a successful prediction model to implement real-time transit discovery in Twitter.