Metadata Induction on a Dutch Twitter Corpus. Initial Phases
Year: 2015
Type: article
Abstract: In this paper, I pose that metadata induction for the TwiNL collection of Dutch tweets should start at a fundamental level, for which I address three types of classification. Identification of users tweeting predominantly in Dutch is shown to be possible with an F-value over 98%. For the identification of individual humans (as opposed to e.g. groups or bots), no classifier is tested, but a number of useful text-based measurements are presented. Finally, the identification of children of school-going age (by far the largest user group in the collection) is shown to be possible on the basis of just unigram counts with high accuracy (93.5%), increasing to very high accuracy (97%) when at least 1,000 tweets are available.
Author: Hans van Halteren
Institution: Radboud University Nijmegen
Language: English
Subfield: Artificial Intelligence
Field: Computer Science
Domain: Physical Sciences
Open Access status: diamond
OpenAlex ID: W2289515300