A informação veiculada em quadros e gráficos estatísticos é imensa e multivariada. Geralmente, fazem-se análises particulares segundo o ângulo que ao observador interessa mais, que noutros momentos, pode ser muito diferente. Assim, creio que o disponibilizarem-se quadros e gráficos estatísticos aos estudiosos é muito importante e que, num blogue, permite que se faça debate das interpretações e análises que cada observador faz dos dados apresentados.

quarta-feira, 18 de novembro de 2009

Weekends and Afternoons Show the Highest Twitter CTRs

via Dan Zarrella by Dan Zarrella on 10/27/09

Want more clicks? My new data suggests that you should Tweet your links in afternoons, evenings and on weekends.

Continuing the study of Twitter clickthrough rates I started last week, I added over 100 more of the most followed Twitter accounts to my database and indexed click data on over 20,000 bit.ly links Tweeted by those accounts. In all of the data below, I measured CTR as the number of clicks a link received, divided by the number of followers the sending account had on the day it Tweeted it. As I noted in my other post, this number can be over 100% due to ReTweets that may use the same bit.ly link.

The graphs below shows the percentage of difference in CTR at each hour or day from the specific average for each account. I did it this way to account for the wide variation in CTRs between accounts (some accounts have much higher rates than others).

The first data point I analyzed is time of day (EST). It showed the expected afternoon/evening preference seen in my other Twitter stats.

Next I looked at days of the week, which showed a much less expected weekend preference. I believe this is due to the "link fatigue" present during the weekdays, where there is a much higher level of activity and many more links are posted.


Download the Science of ReTweets Report here.

Modeling ReTweet Dynamics

via Dan Zarrella by Dan Zarrella on 10/26/09

Earlier this year I read a paper called "Modeling Blog Dynamics" in which they propose a method of modeling the spread of links through the blogosphere using zero-crossing random walks and exploitation vs. exploration applied to a logical flowchart model:

The authors suggested that the model could be used in influence maximization algorithms which aim to identify key, influential individuals in a given social network for the purposes of viral marketing. I was intrigued by the possibilities and have been tossing around a possible flowchart model of how individuals decide to ReTweet specific Tweets since reading that paper. Here's my first attempt:

There are three steps in the process where a marketer can increase the chances of a specific Tweet being ReTweeted. The first step indicates that a user must be following the sender of the target Tweet; the second step means that they must actually see the Tweet in question (try to imagine what percentage of your friend's timeline you actually see). Step three is where the user must find some motivation to ReTweet it.

Maximizing the number of followers the Tweet's original sender has is fairly straightforward, and most of my Science of ReTweets data has explored the ReTweet motivation percentage. I had not put much effort into analyzing statistics around the attention problem, but I've begun to.

Because there is no way to exactly measure what percentage of followers will actually read a given Tweet, the next best metric we have is click through percentages, so that is what I've been working with. You can expect to see more work to that end in the next few weeks.

My work has been concentrated on maximizing the contagiousness of ideas, whereas much of the aforementioned academic work focuses on the people involved in spreading ideas. So you can also expect to see me advance the concepts of "ReTweetability" I began a few months ago with the purpose of identifying influential users.


Download the Science of ReTweets Report here.

Want More Clicks? Tweet Less

Um livro interessante, de apenas 22 páginas, que apresenta dados estatísticos dos utilizadores do tweetter e que nos mostra as tendências comportamentais destas pessoas no fazerem re-tweets. O estudo responde, em parte, à seguinte questão - Qual a melhor hora para se fazer um re-tweet de modo a ser lido por um maior número de pessoas?
http://danzarrella.com/science-of-retweets.pdf
Rui Moio

via Dan Zarrella by Dan Zarrella on 10/21/09

Tweet Much? Don't Expect a High CTR. New data I've been working on seems to indicate that the more frequently you Tweet links, the fewer clicks you'll get.

I've been working towards a statistical model of how an individual makes a decision to ReTweet a specific Tweet and in that process, I came across an interesting problem: before someone ReTweets something, they have to notice it. If you're anything like me, you're only able to actually read a small percentage of the total activity in your friend's timeline, which means that very few of the Tweets I'm technically "exposed" to ever even have the chance of being ReTweeted.

As a measure of "attention," I started looking into click-through data. The wonderful thing about bit.ly is that it has an API that allows anyone to view the stats on any bit.ly link. I grabbed as many of the bit.ly-containing Tweets of several of the most followed and link-heavy Twitter accounts as the Twitter API allows (it imposes a limit of 3,200 total Tweets accessible per user) and the number of clicks each link had gotten. For the time of each Tweet, I also pulled the number of followers that account had and calculated a followers-to-clicks conversion rate. I'll call this rate CTR for simplicity's sake. I was able to get this information for about 2000 Tweets. It is important to note that ReTweets of a bit.ly containing Tweet (if the ReTweeter does not change the link) also count toward the total number of clicks, so it is possible in some cases for a link to have a CTR of over 100%.

Digging into this data, I started to notice an interesting trend: the higher the number of links an account Tweets in a given timeframe, the lower the CTR on each individual link. If you want your Tweet to get noticed and ReTweeted, you should slow down your posting rate.

First, I looked at this data hourly, by graphing the CTR of Tweets over the number of other Tweets posted in the same hour. The first graph below shows individual lines for each account measured; the second graph shows an average for all those accounts.

Then I looked at the numbers by day. The CTR fall-off in these graphs seems to be slower than those above, but the trend is still prominent.

I've got a bunch more stats and analysis to run on this dataset to isolate some factors that lead to increased CTR, and therefore increased attention. I'd also love your feedback on data points you'd like to see.


Download the Science of ReTweets Report here.

sábado, 14 de novembro de 2009

TinEye and plugins

via The Idee Blog by Leila Boujnane on 11/13/09

Picture 3

We spend a significant amount of time in the Ideeplex looking over and analyzing data. We work with extremely large data sets (images typically) and it is always interesting to see what rises up to the surface once you dive in.

Our image search engine TinEye is used by a lot of people and it is interesting (for us) to see the browsers used to access TinEye. Since we launched a TinEye Firefox add-on, we received a lot of requests to develop plugins for other browsers particularly Opera, Safari and Chrome. However the bulk of our visitors are Firefox and IE users. Looking at the data our plugin development plan is pretty wrapped up! We can now go and focus on other things (such as TinEye APIs!).

Incidentally the TinEye Firefox add-on is getting close to 400,000 downloads! I want to see the 1 million download before the year end…