Collection of Road Traffic Incidents in Bangkok from Twitter Data based on Deep Learning Algorithm

Abstract

Text processing technology from Twitter to report notification formats that are known in many countries with verification on different languages. This research presents the development of a neural network memory learning model. To solve the problem of classifying incidence patterns and identifying severity of incidents from Thai social media messages. For gathering incident data and reporting incidents externally from a single reporting platform by using deep learning models like MLP, CNN and LSTM which is designed by dividing the study into 3 types, including examination traffic incidence identification pattern that can identify the report as general news or traffic reporting Incident Identification Patterns. These include traffic conditions, accidents, disasters, damaged roads, or other than the aforementioned patterns, and the pattern indicating the severity of the incidence consists of normal level, medium level and lane blocking or stationary levels. The results demonstrated the ability of LSTM learning with the best results in incidence detection and incidence pattern identification at 93.44% and 87.40%, respectively, and the CNN method was able to State the severity of the incidence at best, reaching 91.42%.