Use this URL to cite or link to this record in EThOS: https://ethos.bl.uk/OrderDetails.do?uin=uk.bl.ethos.791842
Title: Mining Twitter for crisis management : realtime floods detection in the Arabian Peninsula
Author: Alabbas, Waleed
ISNI:       0000 0004 8503 8728
Awarding Body: University of Bedfordshire
Current Institution: University of Bedfordshire
Date of Award: 2018
Availability of Full Text:
Access from EThOS:
Full text unavailable from EThOS. Please try the link below.
Access from Institution:
Abstract:
In recent years, large amounts of data have been made available on microblog platforms such as Twitter, however, it is difficult to filter and extract information and knowledge from such data because of the high volume, including noisy data. On Twitter, the general public are able to report real-world events such as floods in real time, and act as social sensors. Consequently, it is beneficial to have a method that can detect flood events automatically in real time to help governmental authorities, such as crisis management authorities, to detect the event and make decisions during the early stages of the event. This thesis proposes a real time flood detection system by mining Arabic Tweets using machine learning and data mining techniques. The proposed system comprises five main components: data collection, pre-processing, flooding event extract, location inferring, location named entity link, and flooding event visualisation. An effective method of flood detection from Arabic tweets is presented and evaluated by using supervised learning techniques. Furthermore, this work presents a location named entity inferring method based on the Learning to Search method, the results show that the proposed method outperformed the existing systems with significantly higher accuracy in tasks of inferring flood locations from tweets which are written in colloquial Arabic. For the location named entity link, a method has been designed by utilising Google API services as a knowledge base to extract accurate geocode coordinates that are associated with location named entities mentioned in tweets. The results show that the proposed location link method locate 56.8% of tweets with a distance range of 0 - 10 km from the actual location. Further analysis has shown that the accuracy in locating tweets in an actual city and region are 78.9% and 84.2% respectively.
Supervisor: Not available Sponsor: Not available
Qualification Name: Thesis (Ph.D.) Qualification Level: Doctoral
EThOS ID: uk.bl.ethos.791842  DOI: Not available
Keywords: text classification ; Twitter ; event detection ; location extraction ; P304 Electronic Media studies
Share: