StackSample

StackSample is a subset of Stack Overflow that contains 10% of its questions and answers. It is a textual corpus structured in three tables (questions, answers, tags) allowing you to work on tasks such as the automatic generation of tags, the prediction of votes or the semantic analysis of the text of technical questions.

Download dataset

Size

Approximately 468,000 rows in CSV format spread over 3 tables (questions, answers, tags)

Licence

CC-BY-SA 3.0

Description

‍

StackSample is a textual corpus extracted from Stack Overflow, including approximately 10% of the questions and answers. It includes the titles, contents, scores, scores, dates, and tags of the publications. Organized into three CSV tables (Questions, Answers, Tags), this dataset provides an excellent basis for NLP projects in connection with technical communities and natural language in a Q&A context.

‍

What is this dataset for?

‍

Train models for the automatic generation or recommendation of tags
Predict the success (votes, closings) of questions based on their text
Analyze the evolution of technologies through Stack Overflow trends

‍

Can it be enriched or improved?

‍

Yes, StackSample can be combined with other Stack Overflow datasets (e.g. Python, R) for targeted analyses by language. We can also enrich the examples with embeddings, summaries, or classify by technological field. Further exploration of user behavior is also possible via ownerIDs.

‍

🔎 In summary

Criterion	Evaluation
🧩 Ease of use	⭐⭐⭐⭐⭐ (Clear CSV format, well-structured tables)
🧼 Need for cleaning	⭐⭐⭐⭐⭐ (Low – data ready to use)
🏷️ Annotation richness	⭐⭐⭐⭐✩ (Good – presence of tags, scores, and metadata)
📜 Commercial license	✅ Yes (CC-BY-SA 3.0, with attribution and share-alike)
👨‍💻 Beginner friendly	🌟 Yes – easy to explore and documented
🔁 Fine-tuning ready	🎯 Useful for pre-training Q&A or classification models
🌍 Cultural diversity	⚠️ Medium – mostly English, technical content