Elamur Research⌕
☰
home / papers / 2026 / Exploring Sentiment Analysis Approaches in a Public Agency Security News Dataset

Exploring Sentiment Analysis Approaches in a Public Agency Security News Dataset

Paper · 2026 · Thiago Ruiz Lobo and Claudia Aparecida Martins

Summary

The paper tackles automatic polarity classification of Portuguese news headlines about POLITEC, a forensic and identification agency in the Brazilian state of Mato Grosso. The agency's 2024–2027 strategic plan calls for tracking how local media portrays it. The authors scraped six state news outlets and manually labeled headlines as positive, neutral or negative using criteria set by the agency's strategy team. Because non-neutral items were scarce, they added older headlines chosen through polarity-related keywords.

They crossed three text representations (Word2Vec, GloVe and pre-trained BERT embeddings) with four neural architectures (LSTM, GRU, BiLSTM, TextCNN). This produced twelve models, which they evaluated per class with Stratified K-Fold cross-validation. The BERT-based variants performed best, reaching F1 around 88–90%. A qualitative error analysis using NILC-Metrix sentiment-word ratios suggests that misleading polarity words cause some mistakes.

The work matters because it supplies a rare domain-specific Brazilian Portuguese resource for public security news. The resulting classifier already supports the agency's monthly media monitoring.

Key points

  • Data: 28,572 news items from 2005–2024 were collected. The final labeled set has 8,638 headlines: 7,109 from 2023–2024 plus 1,529 older headlines selected by keywords. The neutral class strongly dominates.
  • Experimental grid: 3 representations (Word2Vec, GloVe, BERT) × 4 architectures (LSTM, GRU, BiLSTM, TextCNN) gave 12 trained models.
  • Best model: BERT+GRU achieved F1 of 90.07% for positive and 84.94% for negative headlines. Neutral F1 was about 98% across all BERT models.
  • Error analysis: On nine example headlines, only BERT+BiLSTM classified every item correctly. Errors were explained with NILC-Metrix positive_words/negative_words metrics, which rely on the LIWC Brazilian dictionary.
  • Main limitation: A single annotator labeled the data, consulting the communication team on doubtful cases.

Our summary of the paper, in our own words — not its text. The work belongs to its authors; read it at the links above.

 Fields, topics and concepts

📚 Shelves 2The field of science this work belongs to (OpenAlex fields).Computer Science · 16,667 worksArtificial Intelligence · 16,194 works

🧭 Topics 1Research subjects and languages the work is about.Sentiment & emotion · 972 works

Cite

Cite as Thiago Ruiz Lobo and Claudia Aparecida Martins. 2026. Exploring Sentiment Analysis Approaches in a Public Agency Security News Dataset. Proceedings of the 17th International Conference on Computational Processing of Portuguese (PROPOR 2026) - Vol. 1.

The paper itself, not this page. BibTeX from the library's record of the work.

@inproceedings{lobo2026exploring,
  title = {{Exploring Sentiment Analysis Approaches in a Public Agency Security News Dataset}},
  author = {Thiago Ruiz Lobo and Claudia Aparecida Martins},
  booktitle = {Proceedings of the 17th International Conference on Computational Processing of Portuguese (PROPOR 2026) - Vol. 1},
  year = {2026},
  url = {https://aclanthology.org/2026.propor-1.87/}
}

References

32 references, 4 of them with a card of ours. ↗ leads to the work, «find» searches for one we have no id for yet.

  1. E. Aydin and Zeliha. 2022. Sentiment analysis about turkish tv series with web scraping. In 2022 International Congress on Human-Computer Interaction, Optimization and Robotic Applications (HORA), pages 1–4. find
  2. E. F. Ayetiran and O. Özgöbek. 2024. An inter-modal attention-based deep learning framework using unified modality for multimodal fake news, hate speech and offensive language detection. In Information Systems, volume 123, page 102378. find
  3. R. K. Ayyasamy, C. Ponnusamy, K. N. Bhargavi, S. Cherukuvada, G. Charles Babu, and D. T. Amutha, S. Gamu. 2025. A hybrid deep learning framework for fake news detection using lstm-cgpnn and metaheuristic optimization. In Sci Rep 15, 41522. https://doi-org.ez52.periodicos.capes.gov.br/10.1038/s41598-… — [doi:10.1038/s41598-025-25311-x](https://doi.org/10.1038/s41598-025-25311-x) ↗
  4. K. Cho, B. V. Merriënboer, C. Gulcehre, D. Bahdanau, F. Bougares, H. Schwenk, and Y. Bengio. 2014. Learning phrase representations using rnn encoder–decoder for statistical machine translation.
  5. J. Devlin, M. Chang, K. Lee, and K. Toutanova. 2019. Bert: Pre-training of deep bidirectional transformers for language understanding. In NAACL-HLT.
  6. P. Dhal, D. Biswas, Pragya, J. Patra, M. A. Mishra, and P Jha Kumar. 2023. Stacked layer based deep learning approach for fake news classification. In 7th International Conference On Computing, Communication, Control And Automation (ICCUBEA), Pune, India, 2023, pp. 1-5, doi: 10.1109/IC-CUBEA58933.20… — [doi:10.1109/ic-cubea58933.2023.10392014](https://doi.org/10.1109/ic-cubea58933.2023.10392014) find
  7. A. Didolkar and P. Lokulwar. 2025. Sentiment detection in financial news: A deep learning approach to extreme and moderate classification. In International Conference on Intelligent and Innovative Technologies in Computing, Electrical and Electronics (IITCEE), pages 1–6. find
  8. P. P. B. Filho, T. A. S. Pardo, and S. M." Aluísio. 2013. An evaluation of the Brazilian Portuguese LIWC dictionary for sentiment analysis. In Proceedings of the 9th Brazilian Symposium in Information and Human Language Technology. find
  9. M. J. Grant and A. Booth. 2009. A typology of reviews: an analysis of 14 review types and associated methodologies. Health Information & Libraries Journal, 26(2):91–108. find
  10. A. Graves and J. Schmidhuber. 2005. Framewise phoneme classification with bidirectional lstm networks. find
  11. Y. B. Gumiel, I. Lee, T. A. Soares, T. C. Ferreira, and A. Pagano. 2021. Sentiment analysis in Portuguese texts from online health community forums: Data, model and evaluation. In Proceedings of the 13th Brazilian Symposium in Information and Human Language Technology, pages 64–72. find
  12. N. S. Hartmann, E. Fonseca, C. D. Shulby, M. V. Treviso, J. S. Rodrigues, and S. M. Aluísio. 2017. Portuguese word embeddings: Evaluating on word analogies and natural language tasks. In Symposium in Information and Human Language Technology (STIL). find
  13. S. Hochreiter and J. Schmidhuber. 1997. Long short-term memory. In Neural Computation, volume 9, pages 1735–1780. find
  14. S. Islam, M. N. Kabir, N. A. Ghani, K. Z. Zamli, N. Zulkifli, M Rahman, and M. Moni. 2024. Challenges and future in deep learning for sentiment analysis: a comprehensive review and a proposed novel hybrid approach. In Artif Intell Rev 57, 62. https://doi.org/10.1007/s10462-023-10651-9. — [doi:10.1007/s10462-023-10651-9](https://doi.org/10.1007/s10462-023-10651-9) ↗
  15. S. E. Leal, M. S. Duran, C. E. Scarton, S. Hartmann, N, and S. M. Aluísio. 2023. Nilc-metrix: assessing the complexity of written and spoken language in brazilian portuguese. Lang Resources Evaluation. find
  16. T. Lobo and C. Martins. 2024. Comparativo de algoritmos de aprendizado de máquina para a classificação de notícias sobre a politec em mato grosso. In Anais da XIII Escola Regional de Informática de Mato Grosso, pages 72–77, Porto Alegre, RS, Brasil. SBC. find
  17. T. Mikolov, K. Chen, G. Corrado, and J. Dean. 2013. Efficient estimation of word representations in vector space. In International Conference on Learning Representations (ICLR).
  18. M. Nascimento, V. Silva, G. Souza, K. Lima, E. Araújo, E. Souza, J. Turet, and V. Carvalho. 2025. Extração de notícias sobre segurança pública para desenvolvimento de corpora em português: uma análise preliminar em cidades do nordeste brasileiro. In Anais do VI Workshop sobre as Implicações da Compu… find
  19. J. W. Pennebaker, M. E. Francis, and R. J. Booth. 2001. Linguistic inquiry and word count: Liwc 2001. Mahway: Lawrence Erlbaum Associates. find
  20. J. Pennington, R. Socher, and C. D. Manning. 2014. Glove: Global vectors for word representation. In Proceedings of the 2014 Conference on Empirical Methods in Natural Language Processing (EMNLP), pages 1532–1543.
  21. O. Prasad, S. Nandi, V. Dogra, and D. Diwakar. 2023. A systematic review of nlp methods for sentiment classification of online news articles. In International Conference on Computing, Communication and Networking Technology. find
  22. B. P. Prathaban, Subash R., B. Sujin, and G. Purushothaman, K. E. Lakshmi. 2025. Precision driven sentiment and topic classification of news articles using indicbert model (ibm). In 2025 8th International Conference on Circuit, Power Computing Technologies (ICCPCT), Kollam, India, 2025, pp. 962-967,… — [doi:10.1109/iccpct65132.2025.11176462](https://doi.org/10.1109/iccpct65132.2025.11176462) find
  23. E. R. M. Seno, L. G. T. Silva, F. S. I. Anno, F. M. Rocha Junior, and H. M. Caseli. 2024. Aspect-based sentiment analysis in comments on political debates in Portuguese: evaluating the potential of ChatGPT. In Proceedings of the 16th International Conference on Computational Processing of Portuguese… find
  24. A. Singh and G. Jain. 2021. Sentiment analysis of news headlines using simple transformers. In Asian Conference on Innovation in Technology (ASIANCON), pages 1–5. find
  25. S. A. Soomro, S. S. Yuhaniz, M. A. Dootio, G. Mujtaba Mujtaba, and J. A. Siffiqui. 2022. Category-based sentiment analysis of sindhi news headlines using machine learning, deep learning, and transformer models. In in IEEE Access, vol. 13, pp. 99985-100001, 2025, doi: 10.1109/ACCESS.2025.3576853. — [doi:10.1109/access.2025.3576853](https://doi.org/10.1109/access.2025.3576853) find
  26. C. Soto, G. Nunes, and J. Gomes. 2019. Avaliação de técnicas de word embedding na tarefa de detecção de discurso de ódio. In In: ENCONTRO NA-CIONAL DE INTELIGÊNCIA ARTIFICIAL E COM-PUTACIONAL (ENIAC), 16. , 2019, Salvador. Anais [...]. find
  27. Porto Alegre: Sociedade Brasileira de Computação, 2019 . p. 1020-1031. ISSN 2763-9061. DOI: https://doi.org/10.5753/eniac.2019.9354. — [doi:10.5753/eniac.2019.9354](https://doi.org/10.5753/eniac.2019.9354) ↗
  28. F. D. Souza and J. B. O. Souza Filho. 2022. Embedding generation for text classification of Brazilian Portuguese user reviews: from bag-of-words to transformers. Neural Computing and Applications, 35:24715–24733. find
  29. Fábio Souza, Rodrigo Nogueira, and Roberto Lotufo. 2020. BERTimbau: pretrained BERT models for Brazilian Portuguese. In 9th Brazilian Conference on Intelligent Systems, BRACIS, Rio Grande do Sul, Brazil, October 20-23 (to appear). find
  30. A. Sreekumar, R. Reshma, and B. Athira. 2023. Comparative study of deep learning models for document classification. In 9th International Conference on Smart Computing and Communications (ICSCC), pages 1–6. find
  31. C. Wang, Y. Li, and Z. Wang. 2023. A novel approach for text classification by combining pre-trained bert model with cnn classifier. In 6th International Conference on Information Systems and Computer Aided Education (ICISCAE), pages 1–6. find
  32. L. Yang. 2022. A brief introduction of the text classification methods. In IEEE International Conference on Electrical Engineering, Big Data and Algorithms. find

Updated

↑ back to top

This page in Markdown → https://md.research.elamur.ai/free/en/papers/lobo-2026-exploring-sentiment-analysis-approaches-public/