Datasets:
annotations_creators:
- machine-generated
language:
- en
language_creators:
- found
license:
- apache-2.0
multilinguality:
- monolingual
pretty_name: stackoverflow_post_questions
size_categories:
- 1M<n<10M
source_datasets:
- original
tags:
- stackoverflow
- technical questions
task_categories:
- text-classification
task_ids:
- multi-class-classification
Dataset Card for [Stackoverflow Post Questions]
Table of Contents
Dataset Description
Companies that sell Open-source software tools usually hire an army of Customer representatives to try to answer every question asked about their tool. The first step in this process is the prioritization of the question. The classification scale usually consists of 4 values, P0, P1, P2, and P3, with different meanings across every participant in the industry. On the other hand, every software developer in the world has dealt with Stack Overflow (SO); the amount of shared knowledge there is incomparable to any other website. Questions in SO are usually annotated and curated by thousands of people, providing metadata about the quality of the question. This dataset aims to provide an accurate prioritization for programming questions.
Dataset Summary
The dataset contains the title and body of stackoverflow questions and a label value(0,1,2,3) that was calculated using thresholds defined by SO badges.
Languages
English
Dataset Structure
title: string, body: string, label: int
Data Splits
The split is 40/40/20, where classes have been balaned to be around the same size.
Dataset Creation
The data set was extracted and labeled with the following query in BigQuery:
SELECT
title,
body,
CASE
WHEN score >= 100 OR favorite_count >= 100 OR view_count >= 10000 THEN 0
WHEN score >= 25 OR favorite_count >= 25 OR view_count >= 2500 THEN 1
WHEN score >= 10 OR favorite_count >= 10 OR view_count >= 1000 THEN 2
ELSE 3
END AS label
FROM `bigquery-public-data`.stackoverflow.posts_questions
Source Data
The data was extracted from the Big Query public dataset: bigquery-public-data.stackoverflow.posts_questions
Initial Data Collection and Normalization
The original dataset contained high class imbalance:
label count 0 977424 1 2401534 2 3418179 3 16222990 Grand Total 23020127
The data was sampled from each class to have around the same amount of records on every class.
Contributions
Thanks to @pacofvf for adding this dataset.