Wikipedia Mining

So I was studying nearest neighbor for my machine learning exam tomorrow and I stumbled across "breadth-first search" and it got me thinking... Okay I've come across it before but never thought of a breadth-first search of wikipedia as a means of finding the nearest neighbour...

I thought about all the internal links that wikipedia keeps and how easy it would be to use each page as a graph node then do a bit of a breadth-first search by visiting all those child nodes. Then I thought maybe the links from the child nodes would be interesting so I created a link counter to keep track of how often a link shows up across all children of a page. The way I implemented it wasn't using the wikipedia api or anything - just scraping the data off the web. This means it kinda downloads a few thousand wikipedia pages for a single query... but still it was interesting!

If I query "Machine learning" the closest match is not "Artificial intelligence" as you might guess, but rather International Standard Book Number. It has 38 links pointing to it as opposed to 30. I looked at a few of the pages and it appears these are links I could ignore - info boxes for journals seem to like linking to the ISBN article.... DOI was up there as well.

Anyhow all these links are taken from all the children of the Machine Learning page on wikipedia. There were 11008 total links only one node away, and of these 6873 were unique.

I thought about adding another level to it, ie exploring the children linked to by the children - but as it already takes a minute or so, I imagine that will take a few hours to run and download a crapload of wikipedia. I minor improvement would be needed if any depth could be specified - we wouldn't want to revisit the same node again and again.

Running the program with input "cheese", I found a few more children - 417 of them. This took about 10 minutes to download all the webpages then it quickly used some sets and dictionaries to work out that cheese is equally related to "Rice" and "Carbohydrate" according to this quick measure,  from all of cheese's internal links 124 of them point to each. Bread is a close follower with 122 links.

Hmm well that was a fun diversion, back to study me thinks.

The code in case anyone wants to play:

#!/usr/bin/env python

from urllib2 import build_opener
import HTMLParser

from HTMLParser import HTMLParser
URL = 'http://en.wikipedia.org'
ignore_urls = [
    '/wiki/Portal',
    '/wiki/Special',
    '/w/index.php',
    '/wiki/Help',
    'wiki/Wikipedia',
    '/wiki/Main_Page',
    '/wiki/Category',
    '/wiki/Template',
    '/wiki/Talk:',
    '/wiki/File:',
    
]

class LinkingHTMLParser(HTMLParser):
    def __init__(self):
        HTMLParser.__init__(self)
        #super(LinkingHTMLParser, self).__init__()
        self.links = set()
    def handle_starttag(self, tag, attrs):
        if tag == "a" and len(attrs) > 1:
            (_,url), (_, title) = attrs[0], attrs[1]

            if url.startswith('/wiki') and not any(addy in url for addy in ignore_urls):
                #print 'Linked to %s at address: %s' % (title, url)
                self.links.add(url)


opener = build_opener()
opener.addheaders = [('User-agent', 'Mozilla/5.0')]

class Page(object):
    '''
    A node on the internet... a webpage

        >>> Page('http://en.wikipedia.org/wiki/Nearest_neighbor_search')
    
    '''
    
    def __init__(self, url):
        self.url = url
        self.handle = opener.open(url)
        self.parser = LinkingHTMLParser()
    
    def __repr__(self):
        return 'Page(%s)' % self.url
    
    def __str__(self):
        return '\n'.join(line for line in self.handle)

    def parse(self):
        self.parser.feed(str(self))
        print 'Number of internal links: %d' % len(self.parser.links)
        
    def create_children(self):
        '''Create nodes for each link in self'''
        self.children = []
        
        for url in self.parser.links:
            child = Page(URL + url)
            self.children.append(
                child
                )
def find_n_most_related_pages(topic='Nearest neighbor search', n=10):
    topic = '/wiki/' + topic.replace(' ', '_')
    p = Page(URL + topic)
    p.parse()
    p.create_children()
    total_links = len(p.parser.links)
    all_links = p.parser.links
    link_counts = {}
    for child in p.children:
        try:
            child.parse()
            #print 'Child "%s" has %d links.' % (child.url, len(child.parser.links))
            total_links += len(child.parser.links)
            all_links = all_links.union(child.parser.links)
            for link in child.parser.links:
                if link in link_counts:
                    link_counts[link] += 1
                else:
                    link_counts[link] = 1
        except:
            print 'error'

    print 'Link counts:'
    for i, link in enumerate(sorted(link_counts, cmp=lambda x, y: link_counts[y] - link_counts[x])):
        if i <= n:
            print '%s has %d links.' % (link, link_counts[link])
    print '*' * 80
    print 'Total links only one node away: %d' % total_links
    print 'Unique pages only one node away: %d' % len(all_links)

find_n_most_related_pages('Cheese')