Empowering Real-Time Traffic Reporting Systems With NLP-Processed Social Media Data

Abstract

Current urbanization trends are leading to heightened demand of smarter technologies to facilitate a variety of applications in intelligent transportation systems. Automated crowdsensing constitutes a strong base for ITS applications by providing novel and rich data streams regarding congestion tracking and real-time navigation. Along with these wellleveraged data streams, drivers and passengers tend to report traffic information to social media platforms. Despite their abundance, the use of social media data in ITS has gained more and more attention as of now. In this paper, we develop an automated Natural Language Processing (NLP)-based framework to empower and complement traffic reporting solutions by text mining social media, extracting desired information, and generating alerts and warning for drivers. We employ the finetuned Bidirectional Encoder Representations from Transformers classification model to filer and classify data. Then, we apply the Question-Answering model to extract necessary information characterizing the reported incident such as its location, occurrence time, and nature of the incidents. Afterwards, we convert the collected information into alerts to be integrated into personal navigation assistants. Finally, we compare the recently posted incident reports from both official authorities and social media in order to provide more complete incident pictures and suggest some open research directions.

Innovating in data-driven production environments: simulation analysis of Net-CONWIP priority rule

“…point-of-sale data (Hartzel and Wood, 2017)) or other types of technology that capture relevant demand data (e.g. RFIDs, natural language processing (Wan et al ., 2020)). Likewise, at each workstation, we may have real-time data on the type of product being manufactured that is captured via sensors and reported in real time via the network.…”

Engaging the Crowd in Sensing for Smart Mobility: A Discrete Choice Experiment

“…Collected mobility data are potentially sensitive, since they could be used to reconstruct information about individual participants, such as commute patterns, routines, or private locations. For example, collection of GIS (Geographical Information System) coordinates simplifies the process of identifying the exact location of drivers, but also increases risks regarding privacy and security [41]. Thus, the risk of re-identification potentially influences the privacy risk as perceived by the user.…”

Big Data for Traffic Estimation and Prediction: A Survey of Data and Tools

“…Deep-learning models have been proposed for conducting information extraction, e.g., an LSTM-CNN was proposed to extract traffic-relevant microblogs in [49], which outperformed the baselines, including the support vector machine model based on a bag of n-gram features. Social media data were further proven effective for traffic accident detection and reporting [50,51], traffic jam management [52].…”