Advancing Multilingual Insights and Cross-Border Policies Based on Entity-Level Sentiment Analysis

The task of sentiment analysis has progressed from a broad document-level approach to a more refined entity-level sentiment analysis (ELSA). The key distinction between these approaches lies in their objectives and the granularity of the analysis. Document-level sentiment analysis seeks to capture the overall sentiment expressed by the author regarding the primary topic of a text, as noted in previous studies. In contrary, entity-level analysis focuses on identifying the sentiments directed at specific entities mentioned within the text, offering a more nuanced and detailed perspective. While sentence-level sentiment analysis interprets the overall sentiment about an entity explicitly referenced in a sentence, entity-level analysis delves deeper to capture sentiments related to particular aspects or attributes of those entities. In traditional document-level sentiment analysis, machine learning (ML) models classify the general sentiment of a text as positive, negative, or neutral. However, modern entity-level approaches leverage advanced pre-trained Transformer-based models, such as BERT, to accurately detect and analyze sentiments for individual entities. This advancement enhances the precision of sentiment analysis, surpassing the capabilities of traditional methods and enabling seamless application across multilingual and multi-faceted datasets. At the same time, the mechanism can be extended to other applications, such as the evaluation of local policies, where specific citizen reports can be analyzed. In this context, sentiment analysis is specialized to attribute sentiment for specific entities or issues. For example, a citizen’s post may express negative sentiment for issues related to transportation, but at the same time positive sentiment for the improvement of road infrastructure. Through the ELSA mechanism, sentiment analysis achieves more detailed and accurate distinction between the entities and topics of discussion, providing clear insights of sentiments at the entity level. Moreover, the experience of applying sentiment analysis is now extended to a multilingual environment, enabling multilingual entity-level sentiment analysis (ELSA) in different subject areas. In applications related to radicalization, sentiment analysis is used to extract emotional metadata from online activities, such as the degree of support for violent actions, by commenting on input data such as tweets. At the same time, in applications related to product marketing, extracting sentiment from online sources allows for a better understanding of consumer reactions to specific products. This information can be used in combination with analytical tools to create and validate models related to marketing strategies, facilitating the understanding of trends and the correlation of emotional changes with the acceptance of products. In addition, it provides the possibility of responding directly to consumer needs through the analysis of their comments. The ability to handle multilingual data and analyze sentiment in multiple languages is one of the most significant improvements, also in alignment with the advent of Large Language Models (LLMs), allowing the scalability of the application in international environments and ensuring high efficiency across a variety of data and domains.