neilkodner.com

Two weeks have passed since Apple’s Co-Founder/CEO Steve Jobs passed away.  Upon his passing, Apple encouraged people to share their memories, thoughts, and feelings by emailing rememberingsteve@apple.com. Earlier this week, Apple posted a site (http://www.apple.com/stevejobs) in tribute to Steve Jobs. According to the site, over a million people have submitted messages. The site cycles through the submitted messages.

I decided to take a closer look at what people are saying about Steve Jobs, as a whole. Looking at how the site updates, it appears to use Ajax to retrieve and display new messages. Using Chrome’s developer tools, I monitored the requests it was making to get the new messages.

Once I found the location of the individual messages, it was trivial to download all of them. The message endpoint URLs are in the format

http://www.apple.com/stevejobs/messages/3679.json?28106802

and a sample message looks like

{
mainText: "This is equivalent to my mom's generation of Elvis dying for me. I am very
sadden and emotionally moved at the moment. He was more influential on my
life than my parents and friends. While my parents loved me and friends
shared fun times. Steve influenced me, motivated me to become the innovated,
creative technologist I have become. I got into computer technology in 1980
and moved to Silicon Valley because of him. I have been one of his biggest
admirers and looked to him as a mentor to push the boundaries of my own
creative abilities to develop technology solutions which I hope made a
difference and impact to the industries I worked in. We've lost a
significant influence and icon in technology. We won't see another person of
his innovation and foresight within my life time. He was the Edison of
technology. He was and is one of my biggest inspirations.

I feel I have lost a close family member"
header: "What Steve Jobs meant to me"
author: "Skip"
location: ""
}

The site makes a request to http://www.apple.com/stevejobs/messages/main.json which returns

 {
 totalMessages: "10975"
 timestamp: "28106802"
 }

So it appears that it cycles through 10975 messages. I didn’t decompose the javascript powering the site to determine this, I just made an assumption. I tried querying values greater than 10975 and they returned 404. I wrote a quick python program to download the messages:

#!/usr/bin/python
import urllib2
import simplejson as json
import time
import codecs

# a page on apple's site shows the # of messages available
# start with 0 and retrieve up to message_range messages
metadata = json.loads(urllib2.urlopen('http://www.apple.com/stevejobs/messages/main.json').read())
message_range = metadata['totalMessages']

# the url for each message. i learned of this url by inspecting
# the network calls to http://www.apple.com/stevejobs
# using chrome's developer tools
url="http://www.apple.com/stevejobs/messages/%d.json"

# create our destination file
# i'm using codecs because it does a better job at handling international characters
output_file = 'stevejobs_tribute.txt'
file_handle = codecs.open(output_file,'w','utf-8')

# helper function to remove tabs and linefeeds
def clean(txt):
  return txt.replace('\n','').replace('\t','')

# iterate from 0 to the max # of messages and download the message text
# for these purposes, I'm ignoring the other fields as they weren't always present
for i in range(0, message_range):
  req = url % i
  data = urllib2.urlopen(req).read()
  data = json.loads(data)
  file_handle.write(clean(data['mainText']) + '\n')
file_handle.close()

So now, we have over ten thousand tribute messages saved to the file stevejobs_tribute.txt. What I was most interested in seeing how many of these messages contain a reference to a certain Apple product.
I came up with a few search terms based on some legendary Apple product names including

• Newton
• Macintosh
• MacBook
• iBook
• Mac
• iPhone
• iPod
• iMac
• iPad
• Apple II family
• OSX
• iMovie
• Apple TV
• iTunes
• LaserWriter (yes, Laserwriter)

products = {'iPhone':{'regex':'iphones?','count':0},
	'iMac':{'regex':'imacs?','count':0},
	'iPad':{'regex':'ipads?','count':0},
	'iTunes':{'regex':'itunes','count':0},
	'iPod':{'regex':'ipods?','count':0},
	'cube':{'regex':'cubes?','count':0},
	'MacBook':{'regex':'macbooks?','count':0},
	'iBook':{'regex':'ibooks?','count':0},
	'Apple TV':{'regex':'apple ?tvs?','count':0},
	'Apple II Family':{'regex':r'(apple )?(2|ii|\]\[|\/\/)([ce\+|]|gs|s)?[^0-9]', 'count':0},
	'LaserWriter':{'regex':'laserwriter?','count':0},
	'PowerBook':{'regex':'powerbook?','count':0},
	'Newton':{'regex':'newton?','count':0},
	'OSX':{'regex':'osx','count':0},
	'iMovie':{'regex':'imovie','count':0},
	'Macintosh':{'regex':'macintosh','count':0},
	'Lisa':{'regex':'lisa','count':0},
	'Mac':{'regex':'mac','count':0},
}

Here’s a screenshot of me testing the Apple II regular expression, using the excellent Regexr.

Overall, out of 10975 messages downloaded(as of now), 2,186, or just under 20% mentioned an apple product by name. Here’s the breakdown of the products mentioned:

LaserWriter        1
iMovie             3
OSX                9
iBook             22
PowerBook         22
Lisa              24
Apple TV          31
Newton            33
iTunes            52
Macintosh        163
iMac             235
MacBook          366
Apple II Family  481
iPad             574
iPod             575
iPhone           875
Mac             1315

More than one out of every ten messages included a reference to a Mac! Nearly one in ten mentioned an iPhone – not bad for a device that’s been out a fraction of the time the Mac has been available.I’m pleased to see so many references to the Apple II including several mentions of the//c, which was my first Apple product.

It’s also interesting to note that out of 33 mentions of Newton, only a handful of those were about the actual Apple product – most were comparing Steve Jobs to Newton himself. Check out my earlier post on NLTK concordance for details on how I did this:

import nltk
import string
f = open('stevejobs_tribute.txt').read()
f = f.translate(string.maketrans("",""), string.punctuation)
foo=nltk.Text(f.split())
print foo.concordance('newton')

result:

op If history misses men like Isaac Newton Graham Bell Galileu Thomas Edison a
mbered though his legacy Now he met Newton Einstein and other geniuses like hi
oday I was one of the few who had a Newton Today I have an iPhone 4 an iPad2 a
oduct that came thereafter from the Newton to the Cube to the iPhone 4S God Bl
with the likes of Edison Garcia and Newton for his impact and vision I wish hi
ntioned in the same breath as Isaac Newton Thomas Edison and Bill Gates The le
 off a tree we are thinking of Adam Newton and Steve Jobs He open new dimensio
Jobs will be missed Da Vinci Mozart Newton Franklin Jobs Nobody is out of plac
ged my life starting with the Apple Newton followed by the iPod and then the i
 sorely missed nbsp Da Vinci Mozart Newton Franklin Jobs Nobody is out of plac
ve dared to Einstein Freud Da Vinci Newton Galileo Darwin among others is prou
embered beside Einstein Pasteur and Newton The world is moving toward his crea
irst Apple Mac I remember the first Newton I willnbspremembernbspSteves creati
e to contact us againnbsp How Isaac Newton and Albert Einstein contributed gre
 world One seduced Eve One awakened Newton and One was in the hands of Steve J
the way you have influenced mine If Newton discovered something as remarkable
rld One seduced Eve second awakened Newton the third one was in the hands of S
lent to Leonardo Da Vinci Sir Issac Newton Albert Einstein and the like He was
t of the caliber of that of DaVinci Newton Pythagorous etc The list can go on
hen people say names like ie Edison Newton and Einstein I guarantee that the n
 Computers” The Apple II Lisa Mac Newton iPod iTunes store iPod Touch iPhone
ember Steve Jobs the way I remember Newton or Einstein I lived with Apple prod
set consultant who bought his first Newton MacBook 170 and all the dozens of o
 br 3 Apples change the world Adán Newton Steve Jobs 19552011 Rest in Peace t
back to the Apple IIGS I also had a Newton Steve Jobs death hurts me personall
ed the world apple to adam apple to newton and apple to steve jobs Steve was a
dam and Eva Second one that wake up newton third one that Steve Jobs create St

Also interesting where the number of mentions to other historical figures in the Steve Job remembrance messages. According to the submitters, Steve Jobs is clearly in some elite company. I don’t know if I’d go so far as to group him with the man who brought automobiles and light bulbs to the masses but hey, we all have our priorities. All counts were determined through a simple grep command piped to wc -l.Here are a few examples:

• Einstein – 70
• Ford – 189
• Edison – 110
• DaVinci – 15
• Bill Gates – 8

Finally, I wanted to see what how people were speaking about Steve Jobs and especially what terms were being used to describe him. There was no point in performing sentiment analysis on this text as all of the texts were not only obviously positive but were also vetted by Apple for content. Using NLTK, I performed part-of-speech tagging on every word in each tribute message and then wrote some code to total the adjectives and adverbs used in the tribute messages.

The most commonly-used adjectives are

('great', 1961)
('steve', 1808)
('many', 1459)
('first', 917)
('sad', 862)
('better', 857)
('such', 727)
('best', 721)
('visionary', 645)
('new', 579)
('more', 556)
('true', 538)
('most', 476)
('creative', 471)
('apple', 435)
('other', 427)
('same', 415)
('good', 412)
('greatest', 376)
('wonderful', 373)
('sorry', 362)
('old', 325)
('brilliant', 283)
('able', 281)
('incredible', 267)
('big', 260)

Humorously, NLTK frequently considered “Steve” to be an adjective. This is likely because it is always followed by the proper noun “Jobs.” A tweet from NLTK expert Jacob Perkins reminded me that machines are dumb and proper nouns should be capitalized. In order to aggregate the counts, I normalized the text by converting to lowercase – I wasn’t interested in nouns, only adjectives so proper nouns didn’t matter to me.
The top adverbs, according to NLTK, were not as interesting, at least to me.

('so', 2220)
('never', 2111)
('not', 1897)
('always', 1798)
('just', 1402)
('now', 1028)
('truly', 989)
('only', 945)
('very', 919)
('much', 908)
('ever', 751)
('even', 743)
('really', 567)
('forever', 508)
('more', 486)
('still', 447)
('well', 398)
('most', 375)
('personally', 352)

And finally, I ran tri-gram analysis, again using NLTK. 

trigrams = defaultdict(int)
nltk_trigrams = nltk.trigrams(text)
for itm in nltk_trigrams:
  trigrams[itm] += 1

As one would expect, the leading trigram was ‘rest in peace‘ with 1838 mentions, 16.7% of all mentions. ‘thank you for‘ was found in 1446 messages, ‘will be missed‘ was found in 827 messages. Other interesting trigrams are ‘thank you steve‘ with 791 mentions and ‘changed the world‘ with 551 mentions.

The full python code and resulting data can be found on github.

#!/usr/bin/python
#nltk.help.upenn_tagset('RB')
from collections import defaultdict
from operator import itemgetter
import re
import urllib2
import string
import simplejson as json

import codecs
import nltk

OUTPUT_FILE = 'data/stevejobs_tribute.txt'

adverbs = defaultdict(int)
adjectives = defaultdict(int)
trigrams = defaultdict(int)

message_has_adjective = False
message_has_adverb = False
message_contains_product_mention = False
messages_with_adjective = 0
messages_with_adverb = 0
messages = 0
messages_with_product_mention = 0

exclude = set(string.punctuation)

products = {'iPhone':{'regex':'iphones?','count':0},
	'iMac':{'regex':'imacs?','count':0},
	'iPad':{'regex':'ipads?','count':0},
	'iTunes':{'regex':'itunes','count':0},
	'iPod':{'regex':'ipods?','count':0},
	'cube':{'regex':'cubes?','count':0},
	'MacBook':{'regex':'macbooks?','count':0},
	'iBook':{'regex':'ibooks?','count':0},
	'Apple TV':{'regex':'apple ?tvs?','count':0},
	'Apple II Family':{'regex':r'(apple )?(2|ii|\]\[|\/\/)([ce\+|]|gs|s)?[^0-9]', 'count':0},
	'LaserWriter':{'regex':'laserwriter?','count':0},
	'PowerBook':{'regex':'powerbook?','count':0},
	'Newton':{'regex':'newton?','count':0},
	'OSX':{'regex':'osx','count':0},
	'iMovie':{'regex':'imovie','count':0},
	'Macintosh':{'regex':'macintosh','count':0},
	'Lisa':{'regex':'lisa','count':0},
	'Mac':{'regex':'mac','count':0},
}

def top_n(dct,n = 10):
	srtd=sorted(dct.iteritems(), key=itemgetter(1), reverse=True)
	for x in srtd[0:n+1]:
		print x

def nltk_concordance(term,text_file):
	f = open(text_file).read()
	# remove punctuation
	f = f.translate(string.maketrans("",""), string.punctuation)
	split_text=nltk.Text(f.split())
	split_text.concordance(term,lines=100)

	# >>> f = f.translate(string.maketrans("",""), string.punctuation)
	# >>> foo=nltk.Text(f.split())
	# >>> print foo.concordance('newton')

def unescape(s):
	"""unescapes html codes"""
	s = s.replace("<", "	s = s.replace(" ", " ")
	# this has to be last:
	s = s.replace("&", "&")
	return s

for line in open(OUTPUT_FILE):
	message_has_adjective = False
	message_has_adverb = False
	message_contains_product_mention = False

	# remove the trailing linefeed and convert to lower-case
	# and remove html control characters
	messages += 1
	data = line.strip()
	data = data.lower()
	data = unescape(data)

	# check for product mentions
	for k,v in products.iteritems():
		if re.search(v['regex'],data):
			products[k]['count'] += 1
			message_contains_product_mention = True

	# if the message contains a product mention
	# increment the product mention counter
	if message_contains_product_mention:
		messages_with_product_mention += 1

# tokenize the sentences using nltk's wordpuncttokenizer
	text = nltk.WordPunctTokenizer().tokenize(data)

# compute trigrams
	nltk_trigrams = nltk.trigrams(text)
	for itm in nltk_trigrams:
		trigrams[itm] += 1

# pos-tag each token. we're interested in adjectives and adverbs
	parts_of_speech = nltk.pos_tag(text)
	# test for adjectives and adverbs, increment the counters
	# when we find one.

	for (word,pos) in parts_of_speech:
		if pos.startswith('JJ'):
			message_has_adjective = True
			adjectives[word] += 1

		if pos.startswith('RB'):
			message_has_adverb = True
			adverbs[word] += 1

	# if the message contains an adverb or an adjective, increment a counter
	if message_has_adjective:
		messages_with_adjective += 1
	if message_has_adverb:
		messages_with_adverb += 1

# output the 25 most frequently-used adjectives and adverbs
n = 25
print "top %s adverbs" % n
top_n(adverbs, n)
print
print "top %s adjectives" % n
top_n(adjectives, n)

print "messages with adjectives: %s" % messages_with_adjective
print "messages with adverbs: %s" % messages_with_adverb
print "total messages with product mentions: %s" % messages_with_product_mention
print "total messages: %s" % messages

# output the top 50 most-common trigrams
n = 50
print "top %s trigrams" % n
top_n(trigrams, n)
srtd=sorted(products.iteritems(),key=itemgetter(1))
for x,y in srtd:
	print "%s\t\t%s" % (x,y['count'])

print
print
# concordance for newton
print "concordance for newton:"
nltk_concordance('newton',OUTPUT_FILE)

Inspired by a tweet from Roger Ehrenberg and my 11-year-old son who’s crazy about Chuck Norris facts, I screen-scraped the contents of http://www.chucknorrisfacts.com. Code and data can be found here.

Using Python and BeautifulSoup, it simply loops through all of the pages on http://www.chucknorrisfacts.com and reads the items displayed on the page.

#!/usr/bin/python
import urllib2, time
from BeautifulSoup import BeautifulSoup, BeautifulStoneSoup

# 674 pages last time I checked. Oddly enough, their pages seem zero-based. Additionally, if you
# substitute an arbitrary number, outside of the range of pages, you'll get data back instead
# of 404. I'm not sure why they're doing this.
for page_num in range(0,674):
	url = 'http://www.chucknorrisfacts.com/all-chuck-norris-facts?page=%d' % page_num
	html = urllib2.urlopen(url)
	soup = BeautifulSoup(html)

	entries = soup.findAll("li","views-row")
	for entry in entries:

		# use BeautifulStoneSoup to remove any HTML-escaped text that BS returns.
		the_quote = BeautifulStoneSoup(entry.div.text,
		                   convertEntities=BeautifulStoneSoup.HTML_ENTITIES).contents[0]

		# print it to stdout. I just redirect the program's output to a file.
		print the_quote.encode('utf-8')
	# be a good citizen and wait a few seconds before visiting the next page
	time.sleep(6)

Output looks like

if the mountain won't come to Muhammad, Chuck Norris will bring it.
if you watch the ring you die in 7 days,if you look at Ckuck Norris you die instantly
in a real zombie apocalypse, Chuck Norris can roundhouse-kick 53,596 zombies dead.
in space no-one can hear you scream.....except chuck norris!
iphone 4? chuck norris has iphone 8
most kids pee their name into snow... Chuck Norris pisses his in concreate...
never say you can'thurt a fly to chuck norris because he will hurt you
new never-before-seen behind-the-scenes shots from Walker Texas Range shows Chuck Norris carrying his truck home after it broke down
no one has ever found where the smurfs live thats  becuase they live  in chuck norrises beard
...

Visit my github for the full dataset (5500 entries).

Canabalt, a ridiculously addicting web/IOS-device game allows one to show off their high scores, and their not-so-high scores to Twitter.

Each of these tweets contains a bit of information – The score represented in meters, the method of death (hitting a wall and tumbling to my death) and the device (iPhone). Other useful information can easily be extracted such as the date/time played and information about the user (name, location, friend count, follower count, etc). Over the next few weeks I aim to see what features, if any, has any influence on Canabalt scores.

The first thing I needed to do was capture the tweeted Canabalt scores. I have a process running on an EC2 micro instance that downloads tweets from the Twitter Streaming API based on certain key words, one of them being canabalt. The process loads each matching tweet into a MongoDB instance hosted on MongoHQ.com.

curl -s -u $TWITTER_USERNAME:$TWITTER_PASSWORD -d @/home/ec2-user/trackingkeywords http://stream.twitter.com/1/statuses/filter.json |/home/ec2-user/mongodb/bin/mongoimport &

Where trackingkeywords is a file containing a comma-separated list of keywords that I track on twitter. Additionally, I left connection details out of the mongoimport command. You’ll need to provide a host, port, database, and collection into the mongoimport command.

I then run some python code to query the MongoDB instance and retrieve tweets mentioning Canabalt, based on a simple regular expression. I’m expecting the tweet to begin with ‘I’ and contain the word Canabalt. Pretty naive but it worked fine. If it’s not a true Canabalt score, I’ll be able to determine in no time. From there, I use regular expressions to extract(for now) the score, the method of death, and the device name.

def canabalt_tweets():

	# connect to MongoDB
	tweets = create_connection(False)

	# regular expression to extract components of a canabalt score
	canabalt_regexp = re.compile(r'I ran (\d{3,7})m before (.*) on my ([^.]+)\.')

	# regular expression to match tweets that begin with I ran and mention canabalt
	regexp = re.compile('^I ran .*canabalt')

	# create a MongoDB cursor(query)
	cur = tweets.conftweets.find({'text': regexp}, {'text': 1})

	# iterate through the cursor. If a tweet fits the pattern, print it.
	for item in cur:
		try:
			(score,death,device) = canabalt_regexp.search(item['text']).groups()
			print ','.join([strip_text(score),strip_text(death),strip_text(device)])
		except:
			pass

Function strip_text() is part of my data tools Bat-Utility Belt and cleans text by removing leading/trailing spaces, crlf, tabs and some other junk.

We now have some comma-separated data in this shape

score,death,device
2860,hitting a wall and tumbling to my death,iPhone
3427,hitting a wall and tumbling to my death,iPad
4496,hitting a wall and tumbling to my death,iPad
3635,missing another window,iPhone
2040,colliding with some enormous obstacle,iPhone
6017,somehow hitting the edge of a billboard,iPhone
8374,knocking a building down,iPhone
2939,hitting a wall and tumbling to my death,iPad
2021,turning into a fine mist,iPad

Now for some more fun – visualization and analysis. This is performed in R because, well, R is awesome. That, and I really need some more practice with R.

To date, I’ve collected just over 1200 Canabalt ‘events’. I will likely turn this into a web app if there’s enough interest.

A couple of summaries:

scores by device type:

      device count mean stddev median   max min range
      iPhone   735 4491   3882 3419.0 36332 102 36230
        iPad   284 4723   3884 4041.5 40630 104 40526
  iPod touch   189 3734   3644 2713.0 28024 102 27922>

scores by type of death:

                                            death count mean stddev median   max  min range
          hitting a wall and tumbling to my death   684 4155   3481 3319.5 36332  102 36230
                           missing another window   243 5898   4981 4486.0 40630  409 40221
                         turning into a fine mist    86 3592   2698 2662.5 16441  614 15827
            colliding with some enormous obstacle    40 4768   4247 3256.5 16933  433 16500
                              falling to my death    37 4176   3160 3619.0 13573  567 13006
                       missing a crane completely    22 2950   1774 2923.5  7883  381  7502
                         knocking a building down    21 3399   2267 2849.0  8374  336  8038
                   not quite reaching a billboard    19 3098   1244 2980.0  5772  444  5328
              landing where a building used to be    17 4804   4970 3631.0 22685 1170 21515
          somehow hitting the edge of a billboard    14 5991   3827 5518.5 13547  566 12981
   just barely stumbling out of the first hallway    13  104      1  104.0   104  102     2
              somehow hitting the edge of a crane     7 5497   4835 4942.0 13275  510 12765
       riding a falling building all the way down     4 4278   2162 4195.5  6993 1727  5266
           completely  missing the entire hallway     1 1046     NA 1046.0  1046 1046     0

And now, in the spirit of killing the almighty ink-data ratio, here are some pictures:

What have we learned? So far, while my data set isn’t altogether that large(1200 events), we might have enough to make some basic observations and assumptions(correction please!). Going into this experiment I thought that iPad players would have generally higher scores. This is because of #1 the larger screen size and #2 players wouldn’t necessarily be playing ‘on-the-go’ as they would be (I know I am) on an iPhone or iPod touch. The iPad has higher median and average scores than the other devices. I’d like to revisit this as I collect more data.

The leading cause of Canabalt death, by far, is hitting a wall and tumbling to one’s death. This surprised me as I thought it would be falling to death – that’s how my Canabalt games seem to end.

I’d like to hear your comments suggestions for new analysis, and most of all, your corrections.  You know who you are and this is how I learn. The data and python/R source can be found on github.

The stack: Twitter Streaming API, EC2, MongoDB, Python, Regular Expressions, R

Things I learned working on this: plyr(group-by and aggregation in R), sorting dataframes in R, couple of new ggplot2 tricks.

The 2010 World Cup proved to be one of the most tweeted about events of 2010.  Through the millions of tweets saved to my local Cloudera CDH3 Hadoop cluster, I wrote a quick pig script to discover the ways that people are celebrating(ok, spelling) goals.  Here are the top few variations of Goal/Gol.  The full results can be found here.  I’m happy to share the Pig code if anyone is interested.

158636 gol
126669 goal
31722 Goal
24735 Gol
19610 GOL
14317 GOAL
4178 gool
2981 ggol
2219 goll
1771 goool
1641 GOOOL
1564 Gooool
1498 Goool
1279 GOOOOL
1188 Goooool
1158 GOOOOOOOL
1124 GOOOOOL
1116 gooool
1075 GOOL

A few rough visualizations, they are in need of an update. The first is a scatter between the length of the GOAL and the number of Os contained. You’ll see that a lot of excited soccer fans like to use the full 140 characters in their celebrations, often using 138 Os. Huge thanks to @johnmyleswhite for the inspiration and the R/ggplot2 help. You don’t even want to see my original version!

Inspired by @peteskomoroch, who was inspired by the frequencies of the length of mentions of KAAAAAAHN! here are the frequencies of the length of GOAL! celebrations found on twitter.

A visualization of the hashtag mentions by country through all of the World Cup Tweets.  Click for full-sized version.

And my personal favorite visualization, the number of tweets mentioning vuvuzela, hourly, during the course of the world cup.

After removing a few stopwords and then clearing out a few other words(nowplaying, lastfm, and the like), here’s what’s left.  The data represents a half-day’s worth of tweets.   I’m sitting on about 90,000 tweets about Kanye and am looking forward to taking the time for some more in-depth analysis.  Huge thanks to @jrlevine and @alexmr from twordsie.com for curating the awesome stopwords list, which I found in their github project.