Constructing Twitter Datasets using Signals for Event Detection Evaluation
View/ Open
Date
2014-09-29Author
Hromic, Hugo
Hayes, Conor
Metadata
Show full item recordUsage
This item's downloads: 1051 (view details)
Recommended Citation
Hromic, H.; Hayes, C. Synergies of Case-Based Reasoning and Data Mining Workshop 22nd International Conference on Case-Based Reasoning September, 2014.
Published Version
Abstract
Twitter is a very attractive real-time platform for research on event detection. However, despite the great amount of interest, datasets suitable for evaluating such methods are not easily available. The two most important reasons for this are Twitter's strict Terms and Conditions for data distribution and the vast amount of Tweets data generated at every minute. In this paper we show a rst exploration of a signal processing method suitable for generating datasets for event detection evaluation. Our proposal is based on the notion of ADSR (attack-decay-sustain-release) envelopes commonly used in acoustics signals modelling and applied to Twitter dynamics such as hashtags usage. We show preliminary results over real-world data that support this idea and the potential of our method for the event detection task itself.
Description
Conference paper (workshop)