ANÁLISE DE SENTIMENTO E PREVISIBILIDADE DE SUCESSO DE NOTÍCIAS PUBLICADAS NO SITE DO IF GOIANO

Abstract

This study analyzes the news published on the website of the Instituto Federal Goiano in order to identify which articles receive the most views and the reasons behind their success on the platform. Web scraping techniques were employed using the Python programming language and the Selenium browser automation library to develop a software robot capable of collecting news data from the website. The Apache Airflow platform was used to automate its execution, which was performed daily. The news articles were classified into categories such as public notices (editais) and selection processes, by analyzing the titles and descriptions of the publications with the ChatGPT chatbot, from the GPT series of LLMs by OpenAI. The evaluation of the number of accesses was conducted using Python and the data analysis library Pandas, which provided valuable insights for understanding user interest patterns and preferences. Data analysis techniques involving reading and writing CSV files, selecting and filtering by object classes and numerical values, handling tabulated and missing data, calculating descriptive statistics and Z-Scores, and visualizing data through charts were applied to extract the most relevant information from the publications. Pretrained artificial intelligence models for NLP tasks, such as Google’s Bidirectional Encoder Representations from Transformers (BERT) and Meta’s Bidirectional and AutoRegressive Transformer (BART), were used to perform sentiment analysis on the news. The goal was to determine the influence of more positive language on the number of views and to generate a new description for each article that would receive higher sentiment ratings from the BERT model. The findings indicate that users of the website show greater interest in news related to public notices and selection processes. Additionally, it was observed that publications containing a single cover image tend to achieve a higher average number of views. Finally, the study identified that the most suitable days for publishing news on the IF Goiano website correspond, approximately, to the first business days of the week.