Automatically Identifying Periodic Social Events from Twitter

Proceedings of Recent Advances in Natural Language Processing, pages 320–328, Hissar, Bulgaria, Sep 7–9 2015. Automatically Identifying Periodic Social Events from Twitter Florian Kunneman Centre for Language Studies Radboud University f.kunneman@let.ru.nl Antal van den Bosch Centre for Language Studies Radboud University a.vandenbosch@let.ru.nl Abstract Many events referred to on Twitter are of a periodic nature, characterized by roughly constant time intervals in between occur- rences. Examples are annual music fes- tivals, weekly television programs, and the full moon cycle. We propose a sys- tem that can automatically identify peri- odic events from Twitter in an unsuper- vised and open-domain fashion. We first extract events from the Twitter stream by associating terms that have a high prob- ability of denoting an event to the exact date of the event. We compare a timeline- based and a calendar-based approach to detecting periodic patterns from the event dates that are connected to these terms. After applying event extraction on over four years of Dutch tweets and scanning the resulting events for periodic patterns, the calendar-based approach yields a pre- cision of 0.76 on the 500 top-ranked peri- odic events, while the timeline-based ap- proach scores 0.63. 1 Introduction As a popular communication channel for both sharing news, experiences, and intentions, Twitter has been found to provide an accurate reflection of many aspects of the real world (Bollen et al., 2011; Zhao et al., 2011). For example, the peri- odicity of daily life can be exposed by visualiz- ing the frequency of hashtags such as ‘#breakfast’ and ‘#goodmorning’ (Preo˜tiuc-Pietro and Cohn, 2013). In addition, real-world events can be auto- matically detected by signaling a sudden rise and fall of word occurrences in tweets (Petrovi´c et al., 2010; McMinn et al., 2013). We propose a sys- tem that can identify periodic events from Twitter: provided with a continuous stream of raw tweets, it returns an overview of periodic social events. Surprisingly, this topic of periodicity has not yet been studied in the context of events mentioned on Twitter, while the identification of periodicity in recurring events has obvious gains for a sys- tem that detects events in the Twitter stream. De- tected periodicity patters can be used to predict fu- ture events before they are referred to on Twitter. For instance, if World Food Day is detected on the 16th of October for a number of consecutive years, it can be expected and put on the calendar for the next year. The rich set of references to the real world made on Twitter make it a suitable platform to mine for periodic patterns in relation to events of any type. At the same time, the non-standard language and large amount of streaming messages make it a challenging task. We facilitate this task by ap- plying an event extraction approach that identifies terms that might represent a social event, and that relates them to a frequently and explicitly men- tioned date of the event. After this first event ex- traction stage, periodicity detection can be applied to …

Content truncated to 500 words