| --- |
| license: mit |
| task_categories: |
| - question-answering |
| language: |
| - en |
| tags: |
| - agent |
| --- |
| # Mathematics Stack Exchange Q&A Dataset |
|
|
| A curated dataset of question-and-answer pairs harvested from Mathematics Stack Exchange (math.stackexchange.com). Each entry contains the question, the top answer (if available), metadata (scores, tags, URL), and a combined Q/A text useful for training and evaluation of question-answering and tutoring models. |
|
|
| ## Dataset summary |
|
|
| - Source: Mathematics Stack Exchange (via the official Stack Exchange API) |
| - Content: High-quality Q&A pairs (question title + body, top answer body) |
| - Formats: CSV and JSON |
| - Typical fields: |
| - `question_id` (int) |
| - `question_title` (string) |
| - `question_body` (string, cleaned of HTML) |
| - `question_score` (int) |
| - `question_tags` (string, comma-separated) |
| - `question_url` (string) |
| - `answer_body` (string, cleaned of HTML; empty if no answer) |
| - `answer_score` (int) |
| - `answer_accepted` (bool) |
| - `combined_text` (string, "Q: <title>\n\n<body>\n\nA: <answer or placeholder>") |
| - `scraped_at` (ISO 8601 timestamp) |
|
|
| Example CSV header: |
| ``` |
| question_id,question_title,question_body,question_score,question_tags,question_url,answer_body,answer_score,answer_accepted,combined_text,scraped_at |
| ``` |
|
|
|
|
| ## How the dataset was generated |
|
|
| The dataset was produced by a script that uses the official Stack Exchange API: |
|
|
| - Requests top-voted mathematics questions (configurable count) |
| - For each question, fetches the top answer (if any) |
| - Cleans HTML from bodies (HTML entities unescaped and tags removed), collapses whitespace |
| - Produces both CSV and JSON exports |
| - Respects rate limits and inserts polite delays |
|
|
| Notes: |
| - The generator script is not included in this repository. If you want to include it, add the script file (for example `scripts/generate_math_se_dataset.py`) and reference it here. |
| - For reproducibility, include the script, the exact arguments used, and the timestamp of generation. |
|
|
|
|
| ## Data quality & known limitations |
|
|
| - Bodies are cleaned of HTML but some formatting (math markup, code blocks, LaTeX) may be simplified or lost; consider preserving LaTeX if you need formula fidelity. |
| - Only the top answer (by score) is captured — accepted vs. highest-scored may differ; both are recorded if available. |
| - Potential biases: |
| - Questions with higher votes and English-language questions are overrepresented if using a top-voted query. |
| - Highly specialized or duplicate content may appear. |
| - Empty `answer_body` means no answer was available at scraping time. |
| - Some fields may contain non-ASCII characters; files are exported with UTF-8 encoding. |
