Autor: |
Valliyammai, C., Devi, S. Sri Vaishnavi, Manikandan, D., Vaishnavi, A. K. |
Předmět: |
|
Zdroj: |
AIP Conference Proceedings; 2024, Vol. 2802 Issue 1, p1-9, 9p |
Abstrakt: |
Among social media, Twitter is one of the effective communication platforms during disasters. Since there is a huge increase of users on twitter, it is easy to extract information about the disasters and its location. Information is extracted from twitter through keywords, so there may be a chance that some irrelevant information may also be extracted. So an effective bursty event detection system needs to be proposed to classify the tweets as disasters vs non-disaster tweets. A bursty event detection system is created to identify the disaster events as quickly as possible. The features are extracted using GloVe model. To handle the imbalancement in the data set, the Synthetic Minority Oversampling Technique is used. The classification is done using various deep learning algorithms such as CNN, Multi-Channel CNN, BiLSTM and Hybrid CNN-LSTM. The deep learning models are tuned with hyperparameter tuning to classify the tweets for identifying the bursty events. Based on the results, high accuracy of 96.25% is obtained with Hybrid CNN-LSTM with hyperparameter tuning. [ABSTRACT FROM AUTHOR] |
Databáze: |
Complementary Index |
Externí odkaz: |
|