Traffic Event Detection from Twitter Using a Combination of CNN and BERT
Gregorius Aria Neruda, Edi Winarko · 2021
Knowing traffic situations in a real-time manner is essential in modern society. There are several challenges to using conventional physical sensors. The rise of social media can be an alternative solution to this problem, as it can be a low-cost but still reliable source of information, one of which is Twitter. This paper proposes a combination of CNN (classifier) and BERT (feature extraction) to detect traffic events using social media data from Twitter. Our experimental results show that using the contextual word embedding BERT helps understand the context in tweets and gives better results than non-contextualized word embedding.