Urgency Detection of Events Through Twitter Post: A Research Overview

Abstract

Indonesia is located in volatile Pacific fire ring and stands as one of the most global disaster-prone areas. The government and public are in constant vigilance of the potential threats. Social media generates large amounts of data from its users and is one of the prime candidates to be used as a disaster monitoring and response system. Several studies have demonstrated the effectiveness of using such an approach for developing disaster monitoring systems. but due to the number of messages generated, filtering is needed to extract useful information from the message. Hence, it's needed to detect an urgency from a message. This review, which was carried out following the PRISMA model, focused on the detection of the urgency of Indonesian Twitter during crisis events. From studies selected, several recommendations are suggested for future endeavor in this topic. Future systems are advised to employ three or four class schemes rather than binary class to gain more nuanced insights into tweet urgency. On-Event training and testing approaches are suitable for frequently occurring events, whereas out-of-event approaches offer broader applicability for real-world scenarios. Both machine learning and deep learning methods, including multimodal approaches, have demonstrated efficacy in urgency detection. Addressing pre-processing challenges, such as abbreviations and slang, is critical, and ensuring consistency in manually annotated data is essential. Additionally, the selection of appropriate word embeddings, such as fastText, IndoBERT, or crisis-specific embeddings, is crucial for achieving optimal model performance.

Cite this work

No public copy is stored for this record, so only the metadata is cited.

Stored only when both papers are in the corpus — 15 citations exist upstream.

Papers that cite the same references — the relational fallback while embeddings are still backfilling.