VSE knjižnice (vzajemna bibliografsko-kataložna baza podatkov COBIB.SI)
  • Normalisation, tokenisation and sentence segmentation of Slovene tweets [Elektronski vir]
    Čibej, Jaka, prevodoslovje, računalništvo ; Fišer, Darja, 1978- ; Erjavec, Tomaž, 1960-
    Online user-generated content such as posts on social media, blogs, and forums, is becoming an increasingly important source of information, as shown by numerous rapidly growing NLP fields such as ... sentiment analysis and data mining. However, user-generated content is well-known to contain a significant degree of noise, e.g. abbreviations, missing spaces, as well as non-standard spelling, lexis, and use of punctuation. All this hinders the effectiveness of NLP tools when processing such data, and to overcome this obstacle, data normalisation is required. In this paper, we present a training set that will be used to improve the tokenisation, normalisation, and sentence segmentation of Slovene tweets. We describe some of the most Twitter-specific aspects of our annotation guidelines as well as the workflow of our annotation campaign, the goal of which was to create a manually annotated gold-standard dataset of 4,000 tweets extracted from the JANES corpus of Internet Slovene.
    Vrsta gradiva - prispevek na konferenci
    Leto - 2016
    Jezik - angleški
    COBISS.SI-ID - 60917346