Portuguese Sentiment Analysis with Open-Source LLMs: Models, Prompts, and Efficient Deployment

2026-04-12 · Source: Paper Index on ACL Anthology · Field: Technology & Digital — Artificial Intelligence & Machine Learning, Data Science & Analytics, Natural Language Processing · Depth: Expert, quick

Summary

A comparative study evaluated 29 open-source Large Language Models (LLMs) and two proprietary models for Portuguese sentiment classification. Researchers tested four prompting strategies: Zero-Shot, Few-Shot, Chain-of-Thought (CoT), and CoT with Few-Shot (CoT+FS). The experiments involved approximately 372,000 inferences across a unified three-class benchmark derived from three public review corpora, totaling about 3,000 instances. This process generated roughly 150M input tokens and 65M output tokens. The findings indicate that CoT+FS generally achieves the best performance for larger models. Crucially, several compact open-source models demonstrated competitive F1-scores with significantly reduced computational costs, making them practical for real-world deployments. The study also identified specific teacher–student configurations suitable for knowledge distillation in Portuguese sentiment analysis.

Key takeaway

For AI Engineers building sentiment analysis systems for Lusophone markets, this research highlights that you can achieve robust performance using compact open-source LLMs. Prioritize CoT+FS prompting for larger models, but also evaluate smaller models for their competitive F1-scores and lower computational overhead. This approach allows for efficient deployment without relying solely on proprietary solutions, potentially reducing operational costs and improving scalability for your applications.

Key insights

Open-source LLMs offer competitive Portuguese sentiment analysis with efficient deployment, especially using CoT+FS prompting.

Principles

• CoT+FS generally improves larger LLM performance.
• Compact open-source LLMs can be computationally efficient.
• Knowledge distillation is viable for Portuguese sentiment analysis.

Method

The study compared 29 open-source and 2 proprietary LLMs on Portuguese sentiment classification using Zero-Shot, Few-Shot, CoT, and CoT+FS prompting strategies across a unified 3,000-instance benchmark.

In practice

• Use CoT+FS for optimal performance with larger LLMs.
• Consider compact open-source LLMs for cost-effective deployment.
• Explore teacher–student configurations for knowledge distillation.

Topics

• Portuguese Sentiment Analysis
• Open-Source LLMs
• Prompting Strategies
• Knowledge Distillation
• Computational Efficiency

Best for: AI Engineer, Machine Learning Engineer, Research Scientist, AI Scientist, NLP Engineer, Prompt Engineer

Related on AIssential

• The Architecture Behind Open-Source LLMs — MoE architectures and diverse training strategies drive rapid progress in open-weight LLMs, optimizing cost and performance.
• The DIY AI Trap: Why Your Sentiment Data Might Be Lying to You — LLMs introduce probabilistic variability and biases that undermine DIY sentiment analysis reliability for CX teams.
• Prompt Engineering for Named Entity Extraction from Portuguese Legal Documents — Prompt engineering with LLMs offers a viable alternative for legal NER in low-resource settings.
• Specializing a Small Language Model for Closed-Domain Portuguese RAG using Knowledge Graph Supervision — Fine-tuning SLMs with knowledge graph supervision significantly enhances closed-domain RAG performance, surpassing larger LLMs.
• The signal is the ceiling: Measurement limits of LLM-predicted experience ratings from open-ended survey text — Input text characteristics limit LLM prediction accuracy more than prompt or model choice.
• CARTE: A Benchmark for Mapping Language Model Knowledge Across France — LLMs exhibit significant knowledge gaps in fine-grained, geographically specific regional understanding, particularly within countries.

Open in AIssential →

Editorial summary, takeaway, and curation by AIssential. Original article published by Paper Index on ACL Anthology.