A carregar...
Publicação
Development and optimization of AI algorithms for the categorization of small texts from social media data related to disasters
| Nome: | Descrição: | Tamanho: | Formato: | |
|---|---|---|---|---|
| 1.25 MB | Adobe PDF |
Autores
Orientador(es)
Resumo(s)
Social media is an internet-based form of communication adopted by billions of people around
the world to share information and make connections. One of the most famous social networks
is Twitter, which is often used to give opinions and report news or situations that have been witnessed by the users themselves. In case of emergencies, some of these situations experienced
by users can be useful to identify critical requests. These requests become more frequent during
mass disasters, so the ability to detect critical requests coming from Twitter can improve rescue
services by optimizing the resources available and helping in the decision-making process during
the occurrence of a certain MCI (Mass Casualty Incident).
The system developed in this work aims to monitor social networks during MCIs, using relevant keywords to find tweets and AI algorithms to categorize tweet texts. By splitting the tweets
into certain categories, it is possible to identify those that reveal urgent information. A correlation algorithm was also included in the system in order to identify possible associations between
emergency call information and tweets.
The BERT model was chosen to address the problem after multiple experiments with machinelearning models since it achieved an average of 85% precision, 88% recall, and 86% F1 score in
categorizing the tweets using the testing set.
The categorization of most tweets without errors, the ability to detect urgent tweets, and the
possibility of correlating tweets with emergency call information make this system a useful tool
that can complement the current methods used to monitor MCIs.
Descrição
Tese de Mestrado, Engenharia Informática, 2024, Universidade de Lisboa, Faculdade de Ciências
Palavras-chave
Aprendizagem automática Monitorização de desastres Processamento de Linguagem Natural Mineração textual Análise de redes sociais Teses de mestrado - 2024