EXPRESSO's AI Sentiment Classifier — Goiás Automates Citizen Feedback Analysis for Its State Service Network

EXPRESSO is Goiás's citizen-service delivery model, spanning seven channels (physical stores, counter service, web, app, bot, kiosks and mail). Feedback from citizens using these channels was reviewed manually by SEAD staff, a process the agency's own team described as "time-consuming and prone to inconsistencies."
SEAD's team (Rafaela M. Rosa, Jessé D. de Souza and Camila do N. Freitas) built and tested a sentiment-classification pipeline: text preprocessing (noise removal, tokenization, stemming, stopword tuning) feeding several candidate algorithms, with Multinomial Naïve Bayes selected for its comparatively stronger performance at flagging negative feedback. The validated model was integrated into the state's central Data Warehouse to power a real-time dashboard with word-cloud visualisations of positive and negative comments, letting SEAD managers monitor sentiment on demand rather than wait for manual reports.
Evidence caveat — the published account, presented at the 2025 Conference on Digital Government Research, does not disclose accuracy figures, the volume of feedback processed, or a before/after comparison against the manual process it replaced, and the exact date the tool went live in production (as opposed to being validated with managers) is not stated. The authors describe the approach as "a scalable and replicable framework" they hope to extend to other public service centres, but that extension has not yet been documented.

Read the full analysis: https://proceedings.open.tudelft.nl/DGO2025/article/view/944