Please use this identifier to cite or link to this item:
|Title:||ELD : Event TimeLine Detection : participant-based approach to tracking events|
|Citation:||Mamo, N. (2019). ELD : Event TimeLine Detection : participant-based approach to tracking events (Master's dissertation).|
|Abstract:||People all over the world watch and talk as events unfold, but what does it take for a machine to truly track an event through this crowd-sourced narration? Event TimeLine Detection (ELD) is a real-time Topic Detection and Tracking (TDT) solution to track events using Twitter with the hypothesis that it takes a deeper understanding of the event’s domain for a machine to describe its evolution thoroughly. We test this hypothesis predominantly on football matches as we propose the novel concept of Automatic Participant Detection (APD) to extract an event’s participants before it even starts. We use the detected participants to track events in the TDT component, in which we contribute a novel feature-pivot algorithm. Seeking to explain developments within events, TDT hands our two novel summarization algorithms not only the corpus they are meant to describe, but also the topical keywords within. Our evaluation demonstrates APD’s potential to discover participants early on while exposing the algorithm’s dependency on well-deﬁned environments. As a result, TDT boosts its event coverage through the newly-found participants. TDT’s understanding of developments and their make-up also leads to improved expressiveness in summaries when compared to authoritative content from Twitter and the mainstream media. In fact, an interview with Paul Doyle, a football writer at The Guardian, reveals that ELD’s improvements shift the focus of event tracking from TDT to summarization. Through ELD and its contributions we show that although machines may not have the same understanding that humans accrue over time, they can gain it. In turn, this comprehension permits them to track events closely.|
|Appears in Collections:||Dissertations - FacICT - 2019|
Dissertations - FacICTAI - 2019
Items in OAR@UM are protected by copyright, with all rights reserved, unless otherwise indicated.