A deep learning approach for detecting traffic accidents from social media data (1801.01528v1)

Abstract: This paper employs deep learning in detecting the traffic accident from social media data. First, we thoroughly investigate the 1-year over 3 million tweet contents in two metropolitan areas: Northern Virginia and New York City. Our results show that paired tokens can capture the association rules inherent in the accident-related tweets and further increase the accuracy of the traffic accident detection. Second, two deep learning methods: Deep Belief Network (DBN) and Long Short-Term Memory (LSTM) are investigated and implemented on the extracted token. Results show that DBN can obtain an overall accuracy of 85% with about 44 individual token features and 17 paired token features. The classification results from DBN outperform those of Support Vector Machines (SVMs) and supervised Latent Dirichlet allocation (sLDA). Finally, to validate this study, we compare the accident-related tweets with both the traffic accident log on freeways and traffic data on local roads from 15,000 loop detectors. It is found that nearly 66% of the accident-related tweets can be located by the accident log and more than 80% of them can be tied to nearby abnormal traffic data. Several important issues of using Twitter to detect traffic accidents have been brought up by the comparison including the location and time bias, as well as the characteristics of influential users and hashtags.

• The paper demonstrates that feature extraction using paired tokens with DBN achieves an accuracy of 85% in detecting traffic accidents.
• The paper compares deep learning approaches (DBN and LSTM) with traditional classifiers to effectively manage high-dimensional, noisy social media data.
• The paper validates tweet-derived detections against official accident logs, highlighting potential for near real-time incident response.

A Deep Learning Approach for Detecting Traffic Accidents from Social Media Data

Overview

This paper explores the application of deep learning techniques for detecting traffic accidents through social media data, specifically focusing on tweets. The paper applies advanced deep learning frameworks—Deep Belief Networks (DBN) and Long Short-Term Memory (LSTM) networks—to analyze and classify over three million tweets over a year from two major metropolitan areas, Northern Virginia and New York City.

Methodology

The authors introduce a sophisticated methodology centered on feature selection and classification. An essential aspect of their approach involves feature extraction, concentrating on both individual tokens and paired tokens from tweet text to improve classification accuracy. The paper leverages the Apriori algorithm to establish associations within the token pairs, which underpin their deep learning models.

Two primary deep learning models, DBN and LSTM, are trained and compared against traditional classification methods like Support Vector Machines (SVMs), Artificial Neural Networks (ANN), and supervised Latent Dirichlet Allocation (sLDA). A dataset comprising accident-related and non-accident-related tweets is constructed, supporting evaluations using multiple accuracy and precision metrics.

Results and Findings

The results indicate that feature selection using paired tokens notably enhances detection accuracy. The DBN model outperforms the other models with an accuracy rate of 85%. It surpasses ANN and SVM by better handling the complexities and nuances of social media data, which often present as short, noisy, unstructured texts. The deep learning models particularly excel due to their ability to process high-dimensional data efficiently.

The validation aligns the tweet-derived data with real-world traffic logs and loop-detector data, showing that approximately 66% of the accident-related tweets could be substantiated by official accident logs. Moreover, over 80% of the tweets corresponded with unusual traffic patterns, suggesting the potential to detect previously undocumented incidents.

Implications and Future Directions

This research highlights significant implications for integrating social media analysis into traffic management systems. The ability to detect traffic incidents in near real-time can enhance emergency response capabilities and optimize traffic flow during unreported accidents. Despite its robustness, the paper suggests that social media data should complement rather than replace traditional traffic monitoring methods due to the inherent limitations and noise within the data.

Future advancements could focus on expanding data source integration to involve various social media platforms and exploring the interactions of non-geotagged tweet data. Moreover, there is potential for developing more generalized models that transfer effectively across different geographic locations, enhancing the utility and versatility of their traffic incident detection system. Utilizing larger datasets and exploring additional deep learning architectures could further refine the model's accuracy and broaden its applications beyond traffic incidents to cover a wide array of transport-related phenomena.