Location based bursty event detection and information dissemination using influencers in Twitter

Area of Science:

• Social Media Analysis
• Crisis Informatics
• Network Science

Background:

• Social media facilitates rapid global information sharing, crucial during crises.
• Identifying influential users and pinpointing crisis event locations are significant challenges.

Purpose of the Study:

• To develop a location-based system for detecting bursty events and identifying influencers during emergencies.
• To enable timely dissemination of disaster alerts through identified influencers.

Main Methods:

• Trained machine learning models, including Bidirectional Encoder Representations from Transformer (BERT), to detect emergency tweets.
• Employed Natural Language Processing (NLP) for tweet location extraction using a location tagger and Named Entity Recognizer (NER).
• Utilized the Louvain-based Harmonic Centrality Algorithm to detect user communities and identify influencers within the network.

Main Results:

• The Valence Aware Dictionary for Sentiment Reasoning + Count Vector + Ensemble model achieved 98% accuracy in detecting emergency event tweets.
• The Louvain-based Harmonic Centrality Algorithm identified user communities with a modularity of 0.75.
• The integrated approach successfully identified influencers for effective information dissemination.

Conclusions:

• The proposed system effectively detects crisis events and identifies key influencers for timely emergency alerts.
• This method significantly aids in reducing disaster risk and enhancing information spread during emergencies.