Institutional Repository
Technical University of Crete
Technical University of Crete
EN
|
EL
| URI: | http://purl.tuc.gr/dl/dias/78F16AAE-40B3-49F9-89CC-950F3B0DD51B | ||
| Year | 2019 | ||
| Type of Item | Diploma Work | ||
| License |
|
||
| Bibliographic Citation | Konstantinos Chasapas, "Detection of disaster events via monitoring of social media", Diploma Work, School of Electrical and Computer Engineering, Technical University of Crete, Chania, Greece, 2019 https://doi.org/10.26233/heallink.tuc.82311 | ||
| Appears in Collections |
In our days, the use of social media platforms has gained a lot of popularity. Alarge number of messages are posted everyday on media, such as Twitter. Thesemessages could possibly be used for monitoring real-world events. However, twitter streams contain a large number of meaningless messages and useless content, which may have a negative effect on the monitoring and successful event detection performance. Monitoring and analyzing this rich and continuous user-generated content can yield unprecedentedly valuable information, enabling users and organizations to acquire actionable knowledge. This thesis proposes techniques for natural disaster detection and report by monitoring social media streams. The main task is to use Twitter as a sensor and investigate the data it provides in order to detect and locate a dangerous natural disaster. The detection of such events is attempted by the employment of methods from Natural Language Processing, Machine/Deep Learning and an innovative hybrid combination of the best approaches among the various choices we analyzed (for text representation: Bag of Words, Global Vectors; for text classification: Naive Bayes, Logistic Regression, Random Forest, Decision Tree, Bagged Decision Tree, AdaBoost, Deep Neural Network, Long-Short Term Memory). Finally, our work proposes a spatio-temporal investigation as part of the whole pipelined system to report detected catastrophic events. As a result, the proposed system has been applied with success to Twitter data from 2012 and 2015. The data were referring to two different past disaster events (Nepal earthquake, Hurricane Patricia) and one event not related to disaster (USA presidential election 2012). This system managed to detect both the place and the time of each disaster and confirmed that there is no disaster in the case of the USA elections.
Copyright © DIAS 2013
