An Approach to the Presumptive Detection of Road Incidents in Cuenca, Ecuador Using the Data from the Social Media Twitter and Spanish Natural Language Processing

The road situation in Ecuador, specifically in the city of Cuenca and its surroundings, can vary according to various aspects related mainly to the weather, or traffic incidents caused due to reckless actions by drivers. Therefore, many times these problems can cause some drivers to be forced to delay their trip or look for alternative routes to reach their destination, others can plan their departure according to the related news they find when browsing social media, which can be somewhat time-consuming and even unlikely to find. For this reason, in this paper, a mobile application is presented whose primary function is to collect publications from social media Twitter from different Twitter accounts that are related to road incidents and to classify them using Natural Language Processing (NLP) and Machine Learning (ML) models such as BERT. Then, we present an approximate location of the presumptive incident on an interactive map which allows each of these tweets to be displayed with a personalized icon that can indicate different types of road incidents and, at the same time, the details of the event that occurred. Two main experiments were carried out to find (a) the model accuracy, where a value of 80% was obtained as a result, which is considered a positive result for road incident classification, and (b) the execution of acceptance tests, where a questionnaire based on the ISO 9126 standard was presented to a group of bikers belonging to the city of Cuenca, obtaining as a result that for most users the application is new (85.8%), efficient (85.7%) and easy to use (64.3%). This research makes it possible to present news of road incidents in a centralized manner, which makes it easier for drivers to stay informed and thus avoid annoying interruptions in their circulation.