Premium
Sentiment‐based event detection in T witter
Author(s) -
Paltoglou Georgios
Publication year - 2016
Publication title -
journal of the association for information science and technology
Language(s) - English
Resource type - Journals
SCImago Journal Rank - 0.903
H-Index - 145
eISSN - 2330-1643
pISSN - 2330-1635
DOI - 10.1002/asi.23465
Subject(s) - event (particle physics) , computer science , sentiment analysis , set (abstract data type) , security token , social media , negativity effect , focus (optics) , artificial intelligence , data mining , cognitive psychology , psychology , computer security , world wide web , physics , quantum mechanics , programming language , optics
The main focus of this article is to examine whether sentiment analysis can be successfully used for “event detection,” that is, detecting significant events that occur in the world. Most solutions to this problem are typically based on increases or spikes in frequency of terms in social media. In our case, we explore whether sudden changes in the positivity or negativity that keywords are typically associated with can be exploited for this purpose. A data set that contains several million T witter messages over a 1‐month time span is presented and experimental results demonstrate that sentiment analysis can be successfully utilized for this purpose. Further experiments study the sensitivity of both frequency‐ or sentiment‐based solutions to a number of parameters. Concretely, we show that the number of tweets that are used for event detection is an important factor, while the number of days used to extract token frequency or sentiment averages is not. Lastly, we present results focusing on detecting local events and conclude that all approaches are dependant on the level of coverage that such events receive in social media.