Análise de sentimentos em acordos internacionais de cooperação energética

Abstract

The present work proposes to carry out a sentiment analysis in the structured sentences of international cooperation agreements signed by the Brazilian government, through artificial intelligence, to validate the formality of writing an international agreement based on the sentiment linked to the text, being it positive, negative or neutral. Sentiment analysis is an area that studies, evaluates and classifies affective opinions and feelings in subjective vocabularies, related to Natural Language Processing (NLP), it can classify and describe the attitude or emotion linked to the lexicon. Sentiment analysis can help identify the quality of the texts according to the emotion passed by them, thus, assessing whether their writing complies with what is expected for the standard formality model of the type of vocabulary that was classified. Therefore, a database was created with 181 international agreements written in Brazilian Portuguese signed between 1990 and 2021 and provided by the ENETRIX web tool made for registration, monitoring and analysis of the Global Energy Diplomacy. To evaluate the sentiment in the sentences of international agreements, an artificial intelligence model focused on NLP was used, this model was based on the architecture of BERT neural networks. The model used for sentiment analysis was the "FinBertPTBR: Financial Bert PT BR", which classifies the sentence between Positive, Negative and Neutral feelings. In addition to the analysis of sentiments in the sentences of the international agreements, an analysis of sentiments (Positive and Negative) was also carried out in each of the words extracted from the international agreements, calling this analysis a word polarity assessment. The polarity of the words was classified by the BabelSenticNet API. As a result, we have a detailed analysis of each sentiment classification in the phrases in the database, creating a unique word frequency for each sentiment, as well as a graphical validation of the phrases that were labeled among the sentiments, along with the polarity labels of the words. With this research, a neutral classification was identified as a result for the predominant sentiment of the vast majority of international agreements in the database, consistent with the impartial way of writing this vocabulary model, however, a significant portion of the phrases found in the international agreements in the database data are of positive classification, confronting the final result of text neutrality, leading to the identification of a thin line between these two feelings in the same sentence in the writing of international agreements made available by the ENETRIX tool. The words that stand out most in the change from neutral to positive sentiment in the phrases evaluated are “Strengthen” and “Sustainable”, which are widely used in the definition of an international agreement focused on energy.