The ALPIN Sentiment Dictionary: Austrian Language Polarity in Newspapers

Proceedings of the 13th Conference on Language Resources and Evaluation (LREC 2022), pages 4708–4716 Marseille, 20-25 June 2022 © European Language Resources Association (ELRA), licensed under CC-BY-NC-4.0 4708 The ALPIN Sentiment Dictionary: Austrian Language Polarity in Newspapers Thomas E. Kolb1, Katharina Sekanina2, Bettina M. J. Kern3, Julia Neidhardt1, Tanja Wissik2, Andreas Baumann3 1TU Wien,2Austrian Academy of Sciences,3University of Vienna Vienna, Austria 1{thomas.kolb, julia.neidhardt}@tuwien.ac.at 2tanja.wissik@oeaw.ac.at 3{katharina.sekanina,bettina2.kern, andreas.baumann}@univie.ac.at Abstract This paper introduces the Austrian German sentiment dictionary ALPIN to account for the lack of resources for dictionary-based sentiment analysis in this specific variety of German, which is characterized by lexical idiosyncrasies that also affect word sentiment. The proposed language resource is based on Austrian news media in the field of politics, an austriacism list based on different resources and a posting data set based on a popular Austrian news media. Different resources are used to increase the diversity of the resulting language resource. Extensive crowd-sourcing is performed followed by evaluation and automatic conversion into sentiment scores. We show that crowd-sourcing enables the creation of a sentiment dictionary for the Austrian German domain. Additionally, the different parts of the sentiment dictionary are evaluated to show their impact on the resulting resource. Furthermore, the proposed dictionary is utilized in a web application and available for future research and free to use for anyone. Keywords: Collaborative Resource Construction & Crowdsourcing, Digital Humanities, Document Classification, Text categori- sation, Information Extraction, Information Retrieval, Statistical and Machine Learning Methods, Tools, Systems, Applications 1. Introduction Computational sentiment analysis is a popular method to harness large amounts of textual data and extract information from them. Two main approaches exist for this. The dictionary approach relies on word lists containing sentiment annotations which can be a categorical label (e.g., negative vs. positive) or a numerical value on a continuous scale (from most negative to most positive). The words in a text are then checked against the dictionary to deduce the sentiment of the text. The machine learning approach utilises manually annotated text snippets as training data to extrapolate the sentiment of new, unseen texts (van Atteveldt et al., 2021). A variety of different algorithms can be employed, from relatively simple support vector machines to neural networks (van Atteveldt et al., 2021; Taboada et al., 2011; Liu, 2020; Siegel and Alexa, 2020). Their performance relies on the availability and quality of annotated training data. Attaining high-quality sentiment annotations on large amounts of texts is resource-intensive: Having the texts annotated by a few individuals is time- consuming; relying on crowd-sourcing is expensive. Therefore, dictionaries are widely used, although machine learning approaches have been shown to out-perform dictionary approaches (van Atteveldt et al., 2021). Resorting to dictionary approaches is often less costly in terms of time and money. While dictionary-based sentiment analysis is comparatively easy to implement and more transparent than, e.g., neural networks, it poses the challenge of finding or creating an appropriate dictionary. Furthermore, it is possible to use sentiment dictionaries to create weak labels. This allows the application of supervised methods …

Content truncated to 500 words