Affiliations: [a] Posgrado en Ciencia e Ingeniería de la Computación, Universidad Nacional Autónoma de México | [b] Instituto de Investigaciones en Matematicas Aplicadas y en Sistemas, Universidad Nacional Autónoma de México | [c] Instituto de Ciencias Nucleares, Universidad Nacional Autónoma de México | [d] Instituto de Ingeniería, Universidad Nacional Autónoma de México | [e] Departament de Filologia Catalana i Lingüística General, Universitat de Barcelona
Correspondence:
[*]
Corresponding author. Alex I. Valencia-Valencia, Posgrado en Ciencia e Ingenierıa de la Computación, UNAM. E-mail: [email protected].
Abstract: Social media platforms, such as Twitter (now X), are a major source of communication. Identifying communicative intentions is useful, as it encapsulates the latent motivations that drive text creation. This intention is also helpful in understanding the message, context, and audience. This study proposes a method for detecting communicative intentions in tweets using Jakobson’s language functions. We constructed a meticulously annotated dataset, drawing from the extensive RepLab2013 corpus. Our dataset underwent rigorous scrutiny by linguistic annotators who analyzed over 12,000 tweets individually. These experts identified the dominant language function within each tweet by employing diverse strategies to ensure precise labeling quality. The outcome demonstrated a noteworthy Kappa agreement score of 0.6, reflecting a strong inter-annotator reliability. Subsequently, these functions were mapped to the corresponding intention categories. We employed logistic regression and support vector machines (SVM) algorithms to classify intention in tweets and explored various pre-processing techniques, incorporating n-grams and bag-of-words representations. Furthermore, we expanded our research using pre-trained large language models, incorporating the latest state-of-the-art techniques in natural language processing.
Keywords: Intention, communicative intention, tweets, language functions, intention identification, n-grams, logistic regression, SVM, deep learning
