Taylor Momsen Did Not Write This Slashdot Headline 192
Hugh Pickens writes "David Carr writes that headlines in newspapers and magazines were once written with readers in mind, to be clever or catchy or evocative, but now headlines are just there to get the search engines to notice. Hence the headline for this story that includes a prized key word for one of the 'Gossip Girls' — just the thing to push this Slashdot summary to the top of Google rankings. 'All of the things that make headlines meaningful in print — photographs, placement, and context — are nowhere in sight on the Web,' writes Carr. Headlines have become, as Gabriel Snyder, the recently appointed executive editor of Newsweek.com, says, 'naked little creatures that have to go out into the world to stand and fight on their own.' In this context, 'Jon Stewart Slams Glenn Beck' is the ideal headline, guaranteed to pull in thousands of pageviews. And while nobody is suggesting that the Web should somehow accommodate the glories of The New York Post's headlines in that paper's prime, some of its classics would still work. 'Remember "Headless Body in Topless Bar," perhaps the most memorable New York Post headline ever? It's direct, it's descriptive, and it's oh-so-search-engine-friendly. And not a Taylor Momsen in sight.'"
So it's time to penalize spam headlines (Score:5, Interesting)
This explains (Score:3, Interesting)
Re:A-freaking-men! (Score:5, Interesting)
Re:Whatever Happened to Tagging and Meta Data? (Score:3, Interesting)
You mean putting Meta Tags on my site containing Taylor Momsen, Justin Beiber, and Hardcore Sluts, when my site is really about C# is not a way to generate a random hit? I mean, couple that with re-naming all the variables in example to code to the celebrities on Entertainment Weekly, and reworking all my functions signatures like so;
sex_CalculateInterest(){...
I mean, how do YOU generate activity on your sites? The honest way is for suckers.
Re:So it's time to penalize spam headlines (Score:4, Interesting)
If they could do that, this problem wouldn't exist in the first place. It's a lot harder to asses content algorithmically than to make a reasonable guess as to what the content is about (good or otherwise). The basic problem with any algorithm to detect something (computer or otherwise) is that people start trying to appease the algorithm, not the thing the algorithm is trying to asses. Want to develop a a way to evaluate teachers? Lets test all the students (that's our algorithm), so the teachers teach to the test, and in the end your data is worthless.
In some respects this is the collision between art and science. Computers do science well, art, not so much. Writing articles is an art, even if the content itself isn't, the skill of making an article catchy or otherwise interesting is really hard to evaluate. The google search algorithm is a good example of what happens when you try and apply science, especially early generation science to and art form. How do you judge the quality of any sort of an article? Early on you pick out key words (what about JS and GB for jon stewart and Glen beck, can I detect those?, you might need a context sensitive language to guess their utility, which is possible but again inefficient), you maybe base your evaluation on the status of the author/publisher (something by thomas friedman in the NYT is probably more relevant to a topic than the random crap I post on a blog), so then you need a system of mathematically describing reputation (good luck dodging a bias there). What's the next pass? How many refeences are made to the article elsewhere is probably good, that's a bit of a messy algorithm performance wise but we'll cope. Next up, you get into (for example) verifiability, that's a hard one to do algorithmically for a large data set, and how do you detect someone trying to screw with your verification algorithm. I'm looking at you Anthony Penis Blair, (that's in reference to his longstanding description on wikipedia which I believe has been fixed), and 'michael jackson is dead' (how do you verify that algorithmically when it's breaking news?). A cursory search turns up http://paidcontent.org/article/419-traditional-ways-of-judging-quality-in-published-content-are-now-useles/ listing the criterion for content as 'crediential, correctness, objectivity, crafstmanship'. We can pretty easily do an ok job on credential, correctness is somewhat harder, objectivity and craftsmanship are a really hard. We could maybe make inroads on objectivity by recognizing different objective sets of data and then trying to (machine) learn whether a piece of data fits in one set or another. Craftsmanship is well outside my area of expertise, how do you evaluate the depth and bredth of an article relative to others?
Even with all that, they're taking second seat to data that can be produced quickly, potentially of lower quality but will attract attention of users. It's not an easy problem to solve, it's not just that it's agorithmically hard to evaluate content on the fly (as any science kid in an arts class will tell you), it's that the audience has moved from wanting a certain type of articles (which print media spent the last 400 years perfecting) to wanting instant access to 'probably' correct information, and they have no great attachment to credientials, partly because we've realized that journalists are largely out to lunch when it comes to complex topics.
The print media guys recognized the first problem, probably in time, but the latter problem too late. They needed to adapt their business and publishing model to have significantly more depth, but not necessarily from in house reporters (basically contract an actual expert on each topic and pair them with a writer, sort of like how news networks bring in experts on everything, but with an actual journalism filter on top of them), and they needed to be willing to say on short notice "we have reports from a single source that michael jackson is in fact dead". Probably the natural alliance here is between p
Re:This goes along with ... (Score:3, Interesting)
Google actually tried that for a while. Sort of. There was a little grey X in a box near the end of each search result that would hide it on the page if you clicked it. Complete with a cute little animation of the result poofing into a cloud and contracting on itself. I used it every time I saw expert-sexchange come up in searches.
It went away a while ago. Presumably somebody wrote a bot to X every search result ahead of their own, then spammed the hell out of it.