Wissenschaftliche Untersuchung von Retweets

  • 929 views
Uploaded on

Quelle: Danzarella.com …

Quelle: Danzarella.com

Die hier vorgestellte Arbeit wurde von Dan Zarella im Jahr 2009 veröffentlich.

More in: Education , Technology , Design
  • Full Name Full Name Comment goes here.
    Are you sure you want to
    Your message goes here
    Be the first to comment
No Downloads

Views

Total Views
929
On Slideshare
0
From Embeds
0
Number of Embeds
0

Actions

Shares
Downloads
13
Comments
0
Likes
1

Embeds 0

No embeds

Report content

Flagged as inappropriate Flag as inappropriate
Flag as inappropriate

Select your reason for flagging this presentation as inappropriate.

Cancel
    No notes for slide

Transcript

  • 1. DanZarrella’sThe Science of ReTweets 2009
  • 2. Contents Why ReTweets Matter 4 Psychology RID Content Types 17 More Followers = More ReTweets? RID Attributes 18Distribution of ReTweets per Follower 5 LIWC Attributes 19 RTpF of Suggested Users 6 Timing ReTweeting & Links Time of Day 20 Link Occurrence in ReTweets 7 Day of Week 21 ReTweetability of URL Shorteners 8 About the Author & Data 22 Lingustics Most ReTweetable Words & Phrases 9 Least ReTweetable Words 10 Average Syllables per Word 11 Readability Grade Levels 12 Word Occurrence & Novelty 13 Parts of Speech 14 Punctuation Occurrence 15 Punctuation Types 16
  • 3. “Ideas shape the courseof history.”-John Maynard Keynes 2009 © Dan Zarrella DanZarrella.com 3
  • 4. Why ReTweets MatterI spend a lot of time working on ReTweets, because I believe them to be one of the most impor-tant developments in modern communications, extending far beyond the Twittersphere.Like “The Matrix” was composed of computer code, the real world is made of infectious infor-mation. Your chair, your desk, the computer you’re reading this on, the food you’ll eat today, themoney you’ll earn: they all began as ideas jumping from person to person. None of it would existif the concept wasn’t contagious.Ideological epidemics have made and lost fortunes, they have saved countless lives and causedhorrific wars, they have birthed and destroyed nations. Clearly the most powerful weaponknown to man would be the ability to create powerful mental viruses at will. The very courseof human history would be at your whim.You don’t spread ideas just because they are “good;” you spread them because of some othertrigger or set of triggers has been pulled in your brain. And that trigger fires the biggest gunever seen.And yet, a reliable, repeatable method of crafting a contagious idea has not emerged.The advent of the web changed how memes spread: it made them spread faster, it exposedthem to more people, and it removed many of the constraints imposed by the limits of humanmemory. But there is one change that dwarfs them all: observability.We can now compare millions of viral ideas to uncover the building blocks of contagiousness.ReTweets may seem like a small idea, and they are in some ways. But that small idea is the firstreal window into how ideas spread from person to person. We can study the linguistic traits, thetopical characteristics, the epidemiological dynamics, and the social network interactions thattake place when a person spreads a meme.Not only can this information help us create more contagious Tweets, but many of the lessonslearned through ReTweets will be applicable to viral ideas in other mediums.For the first time in human history we can begin to gaze into the inner workings of the con-tagious idea. That most powerful force can now be put under our microscope and probed for itssecrets. 2009 © Dan Zarrella DanZarrella.com 4
  • 5. Distribution of ReTweets per FollowerWhen I started thinking about how to get more ReTweets, my first thought wasto get more followers; clearly, more followers would mean more ReTweets. Tocheck this assumption, I looked at ReTweets per Followers (RTpF), the numberof ReTweets per day divided by the number of followers.The graph above shows the distribution of RTpF in the top 9000 most followedusers in my database; I’ve graphed the actual distribution line in blue, with a 30-point moving average over it in black.Here we see that while most users had an RTpF of under 1% in my dataset,some users showed much larger ratios, possibly indicating that there are a classof users who are more “ReTweetable” than others.This means that while users with more followers will get more ReTweets, someusers are able to get lots of ReTweets without lots of followers; their contentmust be more contagious. 2009 © Dan Zarrella DanZarrella.com 5
  • 6. RTpF of Suggested Users Non-Suggested SuggestedTwitter maintains a list of users as suggested people for new users to follow; thepeople on this list gain tens of thousands of new followers every day and areamong the most followed people on Twitter.I looked at 200 suggested users and compared them to the 200 most followedusers not on the list. Since many of the suggested users are the most followedpeople on Twitter, they had a much higher average number of followers. So Icompared the two groups using the RTpF metric.The result is clear: suggested users are far less ReTweetable. I think this is likelydue to the fact that many of the followers gained by those users on the suggest-ed list are new Twitter users and may be less ReTweet-savvy. 2009 © Dan Zarrella DanZarrella.com 6
  • 7. Link Occurrence in ReTweets All Tweets ReTweetsI began to study the content of Tweets to identify traits that are correlated withmore contagious, or ReTweetable, content. The first such trait was the pres-cence of a link.I found that in a random sample of normal (non-ReTweet) Tweets, 18.96% con-tained a link, whereas 3 times that many ReTweets (56.69%) included a link.This means that not only are ReTweets an accepted way to spread off-Twittercontent, the prescence of a link may increase a Tweet’s chances of being shared. 2009 © Dan Zarrella DanZarrella.com 7
  • 8. ReTweetability of URL Shorteners bit.ly ow.ly su.pr is.gd 301.to cli.gs twurl.nl ff.im tr.imtumblr.com blip.fmtwitpic.comtinyurl.com -6% -4% -2% 0% 2% 4% 6% 8% 10% I know that most ReTweets contain a link, but there are hundreds of differ- ent URL shortening services available to help you save space with that link. I analyzed my database of over 30 million ReTweets and compared them to over 2 million random Tweets to find which shorteners are the most (and least) ReTweetable. I calculated how much more or less often each URL shortening service appeared in ReTweets than it did in normal Tweets and presented this value as a percent- age. For instance, in my data 9.28% more ReTweets than random Tweets used bit.ly. I took into account the fact that ReTweets tend to contain more links than average Tweets and normalized the occurrence values. I compared the percentage of occurrence for each shortener in random Tweets to the same shortener in ReTweets, to control for the popularity of services like bit.ly and tinyurl.com. The short, post-Twitter shorteners, bit.ly, ow.ly, and is.gd were all more ReTweetable than the older, longer, tinyurl. 2009 © Dan Zarrella DanZarrella.com 8
  • 9. Most ReTweetable Words & Phrases1. you 11. please retweet2. twitter 12. great3. please 13. social media4. retweet 14. 105. post 15. follow6. blog 16. how to7. social 17. top8. free 18. blog post9. media 19. check out10. help 20. new blog postI compared common words and phrases in random Tweets and ReTweets to findthose words that occur in ReTweets more than they occur in normal Tweets.The word “you,” while very common, seems to occur especially often inReTweets, indicating that if you’re talking to “me,” I am more likely to ReTweetit.Its really not surprising that “Twitter” ranks high, but this is a good reminder thatself-reference is always good for buzz in social media.The words “please” and “please ReTweet” are very ReTweetable (“pleasert” also ranked highly). It’s hard to overstate how important it is to ask for theReTweet when you want it; calls to action work.“New Blog Post” is the common prefix used when a person Tweets about, well,a new blog post to their site. That this ranks so highly tells us that Tweeting yourposts is a very smart thing to do. 2009 © Dan Zarrella DanZarrella.com 9
  • 10. Least ReTweetable Words 1. game 11. well 2. going 12. sleep 3. haha 13. gonna 4. lol 14. hey 5. but 15. tomorrow 6. watching 16. tired 7. work 17. some 8. home 18. back 9. night 19. bored 10. bed 20. listeningWhat about the words that are least likely to get your ReTweets?There are a number of “-ing” verbs, including “going,” “watching” and “listen-ing,” which reinforces my understanding that answers to the “What are youdoing?” question don’t get very many ReTweets.The presence of “sleep,” “bed,” “night,” and “tired” indicate that people oftenTweet “goodnight” style messages, but generally don’t ReTweet them.The relatively informal nature of many of the words on the list including “lol,”“gonna,” and “hey,” show that simple or slang conversation is not ReTweetable.The lesson learned here is that if you’re trying to get more ReTweets, don’t justengage in idle chit-chat or Tweet about mundane activities. 2009 © Dan Zarrella DanZarrella.com 10
  • 11. Average Syllables per Word1.631.621.611.601.591.581.571.56 ReTweets Random Tweets I tested the assumption that simplicity is a vital component of ReTweets (as it has been observed in other viral-content types) and I found that random Tweets have 1.58 syllables per word on average, while ReTweets have an average of 1.62 syllables per word. Longer, higher syllable-count words are typically more com- plex, indicating that ReTweets may be more complex than their less viral coun- terparts. Be sure to notice the scale of the graph above, the difference is small, but it cer- tainly challenged my hypothesis that ReTweets are less-complex. 2009 © Dan Zarrella DanZarrella.com 11
  • 12. Readability Grade Levels6.66.56.46.36.26.16.05.95.85.75.6 ReTweets Random Tweets ReTweets Random Tweets Flesch-Kincaid SMOG I then compared two different types of reading grade level analysis metrics and they revealed that ReTweets, in general, are less “readable” and require a high- er level of education to understand. A Flesch-Kincaid test gave ReTweets a reading grade level of 6.47 years of edu- cation, while random Tweets only required 6.04 years. The similar SMOG test (Simple Measure of Gobbledygook) indicated that ReTweets required 6.13 years of schooling, with random Tweets only needing 5.88 years. Again, take care to notice the scale of the Y-axis. 2009 © Dan Zarrella DanZarrella.com 12
  • 13. Word Occurrence/Novelty100908070605040302010 0 ReTweets Random TweetsAnother characteristic commonly found in viral content is novelty; that is, the“newness” of the ideas and information presented. I created a measure of nov-elty by counting how many other times each word in my sample sets occurred.In the random Tweet sample, each word was found an average of 89.19 othertimes, while in the ReTweet sample each word was only found 16.37 other times.This shows us that while simplicity may not be very important to ReTweetability,novelty certainly is. 2009 © Dan Zarrella DanZarrella.com 13
  • 14. Parts of Speech10%9%8%7%6%5%4%3%2%1%0% Noun, plural Proper noun, singular Verb, 3rd person singular present Proper noun, singular Adjective, superlative Adjective, comparitive Wh-adverb Verb, past participle Verb, base form Verb, gerund or present participle Verb, non-3rd person singular present Verb, past tense Noun, singular or mass Adverb Random Tweets ReTweets Part of speech (POS) tagging is an analysis technique in which an algorithm is used to label each word in a piece of content as a specific part-of-speech–noun, verb, adjective, etc. The graph above shows what percentages of words in each sample were labeled as a specific part-of-speech. It lists only the most interesting parts from the much larger list of POS tags. Interesting points from this data include the noun and 3rd-person heaviness of ReTweets, indicating a subject matter and headline type nature. 2009 © Dan Zarrella DanZarrella.com 14
  • 15. Punctuation Occurrence100%95%90%85%80%75% ReTweets Random ReTweets Random With Colons Without Colons I compared a random sample of “normal” Tweets to a sample of ReTweets and found that 85.86% of Tweets contain some form of punctuation, and an over- whelming 97.55% of ReTweets do as well. Of course, the prevailing ReTweet format includes a colon to better display the original Tweet, but even when ignoring this form of punctuation, ReTweets still contain more punctuation than non-ReTweets (93.42% to 83.78%). 2009 © Dan Zarrella DanZarrella.com 15
  • 16. Punctuation Types90%80%70%60%50%40%30%20%10%0% Colon Period Exclamation Point Comma Hyphen Ellipsis Question Mark Semicolon Random Tweets ReTweets I then analyzed the frequency of specific types of punctuation and found that hyphens, periods and colons are the most ReTweetable punctuation, occurring far more commonly in ReTweets than in regular Tweets, while the rarest mark, the semicolon, is the only unReTweetable punctuation mark. 2009 © Dan Zarrella DanZarrella.com 16
  • 17. RID Content Types12%10%8%6%4%2%0% ReTweets Random ReTweets Random ReTweets Random Emotional Primordial Conceptual I used the two linguistic lexicons to analyze ReTweet content: RID and LIWC. First is the more “Freudian” Regressive Imagery Dictionary (RID). This cod- ing scheme is designed to measure the amount and type of three categories of content: primordial (the unconscious way you think, like in dreams); conceptual (logical and rational thought); and emotional. The graph above shows that ReTweets contain less primordial and emotional content than random Tweets and more conceptual content. 2009 © Dan Zarrella DanZarrella.com 17
  • 18. RID Attributes2.5% 2%1.5% 1%.5% 0% Social Behavior Glory Instrumental Sound Vision Abstraction Behavior Random Tweets ReTweets Looking at specific RID attributes, I saw that social and instrumental (constructive words like build and create) behaviors are ReTweetable, while abstract thought and sensation-based words are not. 2009 © Dan Zarrella DanZarrella.com 18
  • 19. LIWC Attributes3.5% 3%2.5% 2%1.5% 1%.5% 0% Occupation We Media Religion Money Insight Negative Emotions Swears Senses Tentative Self-Reference Random Tweets ReTweets The next analysis I performed used LIWC (pronounced “Luke”). This is a lexicon similar to RID, but based in more reviewed and accepted research and refined over 15 years. LIWC measures the cognitive and emotional properties of people based on the words they use. LIWC analysis shows that Tweets about work, religion, money and media/celeb- rities are more ReTweetable than Tweets about negative emotions, sensations, swear words and self-reference. 2009 © Dan Zarrella DanZarrella.com 19
  • 20. Time of Day (EST)6.00% 6% 5.5%5.50% 5%5.00% 4.5%4.50% 4% 3.5%4.00% 3%3.50% 2.5%3.00% 2% 1.5%2.50% 1%2.00% .5%1.50% 0% 1 PM 2 PM 3 PM 4 PM 5 PM 6 PM 7 PM 8 PM 9 PM 12 AM 10 AM 11 AM 1 AM 2 AM 3 AM 4 AM 5 AM 6 AM 7 AM 8 AM 9 AM 11:00 PM 12:00AM 1:00AM 2:00AM 3:00AM 4:00AM 5:00AM 6:00AM 7:00AM 8:00AM 9:00AM 10:00AM 11:00AM 12:00 PM 1:00 PM 2:00 PM 3:00 PM 4:00 PM 5:00 PM 6:00 PM 7:00 PM 8:00 PM 9:00 PM 10:00 PM 12 11 10 Random Tweets ReTweets I compared the volume of ReTweeting that occurs during the day to the volume of regular Tweets to find that ReTweeting is much more diurnal. While overall Tweet volume peaks during business hours and evening, ReTweeting occurs much more frequently between 3 PM and midnight. If you want to get ReTweets, it makes sense to post your content during these hours. 2009 © Dan Zarrella DanZarrella.com 20
  • 21. Day of Week 19%19.00% 18%18.00% 17%17.00% 16%16.00% 15%15.00% 14%14.00% 13%13.00% 12%12.00% 11%11.00% Sun Sun Mon Mon Tues Tue Wed Wed Thu Thu Fri Fri Sat Sat Random Tweets ReTweets I also compared the volume of activity that occurs on various days of the week. Overall Tweeting activity peaks during the business week, as does ReTweeting activity. Monday and Friday are both ReTweetable days, in that a higher percentage of ReTweeting activity for the week occurs than does regular Tweeting. Friday, however, shows the highest volume of ReTweeting, and Thursday the highest volume of standard Tweeting. 2009 © Dan Zarrella DanZarrella.com 21
  • 22. About the Author Dan Zarrella is an award-winning social, search, and viral marketing scientist and author of the upcoming O’Reilly media book “The Social Media Marketing Book.“ Dan has written extensively about the science of viral marketing, memetics and social communications on his own blog and for a variety of popular industry blogs, including Mashable, CopyBlogger, ReadWriteWeb, Plagiarism Today, ProBlogger, Social Desire, CenterNetworks, Nowsourcing, and SEOScoop. He has been featured in The Twitter Book, The Financial Times, NYPost, The Bos- ton Globe, Forbes, Wired, The Wall Street Journal, Mashable and TechCrunch. He was recently awarded Shorty and Semmy awards for social media & viral marketing.A frequent guest speaker and panelist, Dan has spoken at PubCon, Search Engine Strategies, Convergence ‘09,140 The Twitter Conference, WordCamp Mid Atlantic, Social Media Camp, Inbound Marketing Bootcamp, andThe Texas Domains and Developers Conference. He currently works as an inbound marketing manager at Hub-Spot. About the DataOver the course of 9 months, beginning in December of 2008, I’ve collected over 40 million ReTweets, includingTweets that contain variations of “RT,” “ReTweet,” and “Via.” I’ve also collected a random sampling of over 10million “regular” Tweets that may or may not be ReTweets. The Tweets come from Twitter’s search API and itsstreaming API, both of which I have whitelisted access to. I use these 2 data sources and my own PHP scripts toanalyze and compare the characteristics of both. 2009 © Dan Zarrella DanZarrella.com 22