samadhiweb

smalltalk programming for the web



Parsing StackOverflow Data Dump

15 November 2012

Periodically, the Stack Exchange people publish a dump of the content of all public Stack Exchange sites. I played with it back in 2009 when this started, but have lost what little code I wrote back then.

I just downloaded the Sep 2011 dump. For StackOverflow alone, here are the file sizes:

total 43273296
-rw-r--r--  1 pierce  staff    170594039 Sep  7  2011 badges.xml
-rw-r--r--  1 pierce  staff   1916999879 Sep  7  2011 comments.xml
-rw-r--r--  1 pierce  staff         1786 Jun 13  2011 license.txt
-rw-r--r--  1 pierce  staff  10958639384 Sep  7  2011 posthistory.xml
-rw-r--r--  1 pierce  staff   7569879502 Sep  7  2011 posts.xml
-rw-r--r--  1 pierce  staff         4780 Sep  7  2011 readme.txt
-rw-r--r--  1 pierce  staff    193250161 Sep  7  2011 users.xml
-rw-r--r--  1 pierce  staff   1346527241 Sep  7  2011 votes.xml

Assuming each row is a line by itself, there were more than six million posts as of Sep 2011:

% egrep "row Id" posts.xml | wc -l
 6479788

According to readme.txt in the dump package, the file posts.xml has the following schema:

  • posts.xml
    • Id
    • PostTypeId
      • 1: Question
      • 2: Answer
    • ParentID (only present if PostTypeId is 2)
    • AcceptedAnswerId (only present if PostTypeId is 1)
    • CreationDate
    • Score
    • ViewCount
    • Body
    • OwnerUserId
    • LastEditorUserId
    • LastEditorDisplayName="Jeff Atwood"
    • LastEditDate="2009-03-05T22:28:34.823"
    • LastActivityDate="2009-03-11T12:51:01.480"
    • CommunityOwnedDate="2009-03-11T12:51:01.480"
    • ClosedDate="2009-03-11T12:51:01.480"
    • Title=
    • Tags=
    • AnswerCount
    • CommentCount
    • FavoriteCount

I'm not going to build a DOM tree of 6+ millions posts in RAM yet, so I'll use a SAX handler to parse the thing. First, install XMLSupport:

Gofer new
    squeaksource: 'XMLSupport'; 
    package: 'ConfigurationOfXMLSupport';
    load.
(Smalltalk at: #ConfigurationOfXMLSupport) perform: #loadDefault.

As per SAXHandler's class comment, subclass it and override handlers under the "content" and "lexical" categories as needed:

SAXHandler subclass: #TnmDmSOHandler
    instanceVariableNames: ''
    classVariableNames: ''
    poolDictionaries: ''
    category: 'TNM-DataMining-StackOverflow'

For a schema as simple as the above, the method of interest is this:

startElement: aQualifiedName attributes: aDictionary
    aQualifiedName = 'row' ifTrue: [ Transcript show: aDictionary keys; cr ]

Using a 1-row test set, the following do-it

TnmDmSOHandler parseFileNamed: 'p.xml'

produces this output:

#('Id' 'PostTypeId' 'AcceptedAnswerId' 'CreationDate' 'Score' 
  'ViewCount' 'Body' 'OwnerUserId' 'LastEditorUserId' 
  'LastEditorDisplayName' 'LastEditDate' 'LastActivityDate' 'Title' 
  'Tags' 'AnswerCount' 'CommentCount' 'FavoriteCount')

From here on, it is straightforward to fleshen startElement:attributes: to extract the stuff that is interesting to me.

To count the actual number of records, just keep a running count as each post is parsed, and print that number in the method endDocument. The run took a long time (by the wall clock) and counted 6,479,788 posts, the same number as produced by egrep'ping rowId.

How about Smalltalk time? Let's ask TimeProfiler.

TimeProfiler onBlock: [ TnmDmSOHandler parseFileNamed: 'posts.xml' ]

Btw, saw this comment on HN: "If it fits on an iPod, it's not big data." :-)