Method for classifying aspects of argumentation in Russian-language texts

Бесплатный доступ

Argumentation mining in texts has attracted the attention of researchers in recent years due to a wide range of applications, in particular, in the analysis of scientific and legal texts, news articles, political debates, student essays and social media. Recently, a new task has been set in this area- aspect-based argumentation mining, where an aspect is defined as a property of the object, regarding which the argument is being built. Accounting for the aspects allows, on the one hand, to clarify the direction of the argumentation and understanding of the argument structure; on the other hand, it can be used to generate high-quality and aspect-specific arguments. The article proposes a method for classifying aspects of argumentation in texts in Russian. On its basis we train and study the models for classifying aspects of argumentation using machine learning and neural networks. For the first time, a Russian-language text corpus was formed, including 1,426 sentences and marked by 16 aspects of argumentation, a neural network language model ArgBERT for classifying arguments was built, and Random Forest models were trained to classify aspects of argumentation. The classification performance obtained on the basis of Random Forest models is 0.6373 by F1-score. The developed models demonstrate the best performance for the aspects “Safety”, “Impact on health”, “Influence on the psyche”, “Attitude of the authorities” and “Standard of living” (F1-score is higher than 0.75).

Еще

Argumentation mining, text corpora, neural network language models, machine learning, Random Forest, aspects of argumentation

Короткий адрес: https://sciup.org/143181011

IDR: 143181011   |   DOI: 10.25209/2079-3316-2023-14-4-25-45

Статья научная