metadata

configs:
  - config_name: default
    data_files:
      - split: train
        path: stackoverflow/train_clean.csv
  - config_name: raw
    data_files:
      - split: train
        path: stackoverflow/train.csv

StackExchange Dataset

Working doc: https://docs.google.com/document/d/1h585bH5sYcQW4pkHzqWyQqA4ape2Bq6o1Cya0TkMOQc/edit?usp=sharing

BigQuery query (see so_bigquery.ipynb):

CREATE TEMP TABLE answers AS
SELECT *
FROM bigquery-public-data.stackoverflow.posts_answers
WHERE LOWER(Body) LIKE '%arxiv%';

CREATE TEMPORARY TABLE questions AS
SELECT *
FROM bigquery-public-data.stackoverflow.posts_questions;

SELECT *
FROM answers
JOIN questions ON questions.id = answers.parent_id;

NOTE: BigQuery only has the StackOverflow site data, not the other sites. So if we want to query the other sites, we would probably want to download the data dump to a cluster and run a SQL server.

Columns in the raw query output:

'id', 
'title', 
'body', 
'accepted_answer_id', 
'answer_count', 
'comment_count', 
'community_owned_date', # present only if post is community wiki'd
'creation_date',
'favorite_count',
'last_activity_date', 
'last_edit_date',
'last_editor_display_name', 
'last_editor_user_id',
'owner_display_name', 
'owner_user_id',
'parent_id', # if post is answer, then this is the question id; if post is question, this is None
'post_type_id', # 1 = QUESTION, 2 = ANSWER
'score',
'tags',
'view_count',

(Official database schema)[https://meta.stackexchange.com/questions/2677/database-schema-documentation-for-the-public-data-dump-and-sede/2678#2678]

File structure

Each folder represents the different StackExchange sites (~182). The largest one is StackOverflow.