Computational analysis of Orientalist discourse in social media: A case study of Twitter

Abstract

The paper presents ORIENT-Net as a tool that processes texts in terms of their orientalist views. This is done by the detection and classification of such texts on social media Twitter. An Orientalist view of the East can see the cultures there as being quaint, obsolete, or entirely unlike the West, which Westerners can be very often influenced by. The demographic figures are also computed for the ORIENT-Net project, to be able to provide the necessary NLP, ML, and data mining solutions that are tailored to the specific demographics. The framework collects the relevant tweets on the basis of keywords that also cover a range of different kinds of Orientalist statements. The very subtle orientalist markers are identified through three linguistic features: TF-IDF, n-grams, and BERT embeddings that are the result of fine-tuning. The CatBoost classifier is the one that does the classification of the tweets into ‘orientalist’ and ‘non-orientalist’ categories, and the system’s performance is measured through accuracy evaluation, along with precision and F1-score assessments. The model relies on SHAP and LIME for interpretability and decision explanation. The framework makes online detection of oriental discourse automatic. The results indicate the high accuracy and clarity of orientalist identification. This system offers a scalable tool to support sociocultural research and inform evidence-based policy decisions. The uniqueness of this research is that it combines postcolonial theory and computational modelling via the ORIENT-Net framework, thus providing an interpretable and scalable method for online ideological discourse analysis. Future work will focus on expanding the dataset across multiple languages and platforms, as well as refining detection capabilities to capture more nuanced and context-dependent discourse patterns.