Graph-Based Lexical Sentiment Propagation Algorithm

In the rapidly developing field of sentiment analysis, it is a challenge to create sentiment dictionaries with broad coverage, especially for languages with limited resources. To address this problem, we propose innovative methodologies that automate the creation of comprehensive sentiment dictionaries, utilising both traditional linguistic approaches and state-of-the-art artificial intelligence technologies. The methodologies are characterised by their universal applicability to different languages. The proposed ConGraCNet Sentiment Propagation algorithm uniquely combines existing sentiment dictionaries and corpus-based syntactic–semantic embedding graphs to reliably capture and propagate sentiment values in lexical networks. To demonstrate the particular benefit for underrepresented languages with scarce sentiment resources, such as Croatian, we used the ConGraCNet Sentiment Propagation algorithm to create the Sentiment-hr dictionary and the AI tool GPT to generate the Sentiment-hr-AI dictionary. The two open-source sentiment dictionaries created are the largest and most comprehensive resources for the Croatian language to date, being at least ten times larger than the second-largest sentiment dictionary available. Our results demonstrate the effectiveness of the methods presented, which significantly expand the toolkit of sentiment analysis for the Croatian language and provide researchers with valuable insights and resources.