george.james's blog

I think it was Rob who told me about how spammers are using porn sites to decode Captchas.

Captchas are those fuzzy images containing words, letters or numbers that you have to type in on web-sites to prove that you are a human being. The idea is to make it more difficult for bots to automatically sign up for sites where they can inject spam messages or, more usually, web-site links that then help to improve the target site's Google ranking.

So the clever spammers have started grabbing the Captchas from these sites and putting them in front of porn sites. The porn site users are only too eager to decipher the Captcha so that they can get to see the goods. The spammers then immediately use the result to get access to the original site.

While there is some evidence that this might just be an urban myth, it is a smart implementation of the principle that I mentioned during my talk about how web-sites can glean small fragments of data from large numbers of users and then aggregate it into large useful datasets. Amazon, with their Mechanical Turk have industrialised the whole principle.

So, today I came across another interesting application of Captchas. The School of Computer Science at Carnegie Mellon has developed a service called reCAPTCHA. It uses words that have been taken from scanned books that cannot be deciphered by conventional OCR techniques. The results are passed back to reCAPTCHA where they are assembled to create a digitized version of the original text. Their strap-line Digitizing Books One Word At A Time says it all really.

Two words are shown for each Captcha, one of which has a known answer and the other unknown. The user doesn't know which is which so the have to type in both words. If the known word validates correctly then it is assumed that the unknown word has also been entered correctly. Once an unknown word has been deciphered correctly it can be used as a known word in the future. The whole thing is extremely clever and well thought out.

I especially like the way that it provides a protection service for web-sites while at the same time performing a valuable service that digitizes books from the The Internet Archive. A real win-win service.

We asked delegates to fill out a feedback form. We kept it simple, there was only one question: How was it?

Thank you to everyone who took the time to tell us what you thought.

Here is the best of them:

And here are the worst of them:

Full disclosure: These feedback comments are unedited, except we have cropped those where people signed their names, and cropped one (positive) sentence that would have disclosed an author's identity. All feedback forms completed at the conference have been included above. A few comments were also received subsequently by email, but I can't find where Michelle has filed them. I believe they were all positive.

Podcasts of all the main presentations are now available.

Thanks must go to Simon Tweed who has tirelessly processed the audio recordings, cleaned them up, sliced and diced them and done all kinds of post-processing to make them into mp3s. The end result is exceptionally good - much better than the kind of quality I'm used to from conference recordings. They are a joy to listen to.

Each session is a separate mp3 file. You can listen to it live on the site or download it to your mp3 player and listen at your leisure.

The presentation slides are been hosted by Slideshare who provide a nice flash based applet that allowed them to be view directly on this site. As an aside, slideshare uses Amazon S3 for its storage and content hosting - S3 was much discussed during the conference.

A nice feature of slideshare is that the audio tracks can be synchronised with the slide presentations. As time permits we will be doing this for some of the presentations. The first, Querying the Web, has already been done. If you want the best possible viewing experience then click the little green play button beneath the slides, sit back, relax and enjoy.

Oh, and one last thing. The nice people at slideshare have picked my Querying the Web presentation as one of their featured slidecasts, which is really cool.