Automatic Question classification in Portuguese: A Large-Scale Dataset and Comparative Evaluation of Classification Strategies

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

Summary

Murilo Boccardo and Valéria D. Feltrim presented a comparative evaluation of automatic classification strategies for Brazilian university entrance exam questions, categorizing them by subject and fine-grained topic. Their central contribution is a new large-scale Portuguese-language dataset, curated from approximately 17,000 questions from the Agatha.edu platform, which were meticulously cleaned and normalized. The study explored two classification approaches: a single-step method directly predicting fine-grained topics and a two-stage method involving initial subject prediction followed by specialized topic classifiers. These strategies were assessed using both classical machine learning techniques, including Support Vector Machines, Naive Bayes, and Random Forest, and transformer-based language models pre-trained for Portuguese. Experimental results confirm the viability of large-scale automatic question classification and underscore the utility of NLP-based methods for managing educational question banks.

Key takeaway

For research scientists developing educational technology in Portuguese-speaking markets, this work demonstrates the feasibility of large-scale automatic question classification. You should consider adopting a two-stage classification strategy, combining classical machine learning with transformer-based language models, to accurately categorize exam questions by subject and fine-grained topic. This approach can significantly improve the curation and organization of your educational question banks.

Key insights

A new 17,000-question Portuguese dataset enables large-scale automatic classification of exam questions.

Principles

• Two-stage classification can refine topic prediction.
• NLP strategies enhance educational question bank management.

Method

The study compared single-step fine-grained topic prediction against a two-stage approach: subject prediction followed by specialized topic classifiers, using both classical ML and transformer models.

In practice

• Use transformer models for Portuguese question classification.
• Implement a two-stage classification for granular topics.

Topics

• Automatic Question Classification
• Portuguese Language Dataset
• Educational Question Banks
• Transformer Language Models
• Machine Learning Classification

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

Related on AIssential

• Automatic Evaluation of Enem Essays: A Comparative Analysis between Feature Engineering and Transformers — Regression formulations are generally more suitable for ordinal essay scoring than multiclass classification.
• LARI Dataset: A Native Portuguese Question Answering Dataset from Brasileiras em PLN — LARI dataset provides a human-validated Portuguese QA resource, addressing a critical language resource gap.
• NCTB-QA: A Large-Scale Bangla Educational Question Answering Dataset and Benchmarking Performance — Domain-specific fine-tuning is crucial for robust QA performance in low-resource languages, especially with unanswerable questions.
• Automatic Reflection Level Classification in Hungarian Student Essays — Automated reflection classification in Hungarian essays is feasible using both classical ML and fine-tuned transformers.
• Deep text-pair classification with Quora's 2017 question dataset — Quora's 2017 dataset of 400,000 question pairs enables deep learning solutions for text-pair semantic equivalence tasks.
• Improving Answer Extraction in Context-based Question Answering Systems Using LLMs — Fine-tuning pre-trained LLMs on benchmark QA datasets significantly improves answer extraction accuracy and contextual understanding in QA systems.

Open in AIssential →

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