I picked up a copy of The Wall Street Journal on Saturday and did something I hadn’t done in many years – look at the stock tables. I’d been on vacation for a week without my laptop and I wanted to know how IFN was doing.
The WSJ, like most papers, organizes their stock tables by exchange. I was pretty sure that IFN is traded on the NYSE. No luck. I looked through NASDAQ, AMEX, and a bunch of other tables. I gave up after 2 or 3 minutes.
The experience reminded me of the conflict between search and browse experiences. Think of “NYSE”, “Nasdaq”, and “AMEX” as tags or categories. I could browse through the categories, but why?
It requires me to think the same way as the person who assigned the tag or category. On most financial Web sites, I can just type in “IFN” and get the right answer without thinking about the appropriate classification.
I’ve written before about how local search forces people to think like computers. It’s easy to assign a ZIP code using rules that the post office has created. Unlike tagging, it’s also consistent. (Rich Skrenta has an excellent post on problems with tagging.) But it isn’t the best way for humans to find a place.
Classification systems force information seekers to think like the person who designed the system. In many cases, those designs are arbitrary and inconsistent. A former Yellow Pages salesman told me that if a potential advertiser didn’t agree with a category, he’d just create a new one to win the sale.
If I want to find the “Irish pub south of Grand Central” I should be able to do that without knowing how Irish pubs are categorized (are they nightlife? bars? restaurants? entertainment?) or what the ZIP code for the area is.
In the case of IFN, I’ve got money riding on it, and I still wasn’t 100% certain which tag would get me my answer.
Rakesh Agrawal's blog 
Very right Rocky. You rock! I can’t stand thinking like a computer either, boring and buggy. Categories made sense because there was no better way. Our brains don’t start a thought with some category disambiguation, very right. I am no anthropologist, although it feels like I think along concepts and associations, no directory categories and then drill down a few linear subcategories to reach my hypothalamus. Concepts are multidimensional, not a category 1 >> subcat 4 >> sub-subcat 3 and so forth. It’s like a giant multidimensional sphere of knowledge where concepts are related or not, at a certain distance of each other. Navigate that!
-arnaud
> Classification systems force information seekers to think like the person who designed the system.
This is correct if, and only if, you stick the classification system into the user’s face, and order “you must browse via the hierarchy and nothing else”.
But – and this is a point that the mainstream critizism of classification systems keeps missing – while this is pretty close to the reality that seekers face today when using the models for classification of web content, the Yahoo directory and Open Directory, this reality is far away from the real potential of these classification systems. They were put into deep-freeze in an early, and rather imperfect, stage of their development, with datamodels that allow only limited development, with simplistic display and search functions. And yes, because of these limitations the user is forced to think like the person who created the hierarchy – but this is not the fault of the classification systems themselves, it is the fault of the miserable last-century technology and usability in which they got caught up. Imagine you set up a little school library, and some years later someone decides that the same simple classification system and small shelves and manually-written catalogue cards are to be used for a large modern university library: of course it won’t work, it can not work – but not because the idea to organize books in a library is wrong in principle.
And… talking of the real potential of classification systems… there’s a second point that tends to get overlooked: while most commentators praise the advantages of algorithmic search and tagging over hierarchic classification, the R&D departments of the big search engines are busy experimenting how the output of the despised old classification systems can be used to enhance algorithm-based search. E.g. to offer choices via various “related topics” features. Or for ranking purposes in the algorithms, invisible and unnoticed by searchers. Below the hood of the search engines, data created via classification, and the taxonomy of the classification systems itself, can even be used to address the very problem that you complain about: that you have to enter the right keywords into the search engine, pretty much the same way as you have to guess the correct path when browsing in a directory, or you may never get the results which fit best for your needs. ( example; example; more examples )
The old antagonism “either hierarchic classification or algorithm-based search” doesn’t make sense anymore under these circumstances – when the search engine that you prefer to a directory relies on that same directory to improve your search experience. What you should ask is not: how to get rid of classification – the question that needs to be asked is: how can we get the classification systems out of the corner in which they were forgotten years ago, revive them, improve accessability of their content, replace their black-sheep image by something that fits better to their actual importance for the future of search, and tap their full potential to create data that can power next-generation search tools.
Chris,
You’re absolutely right. My post was in reference to the navigation value of classification systems. The effectiveness and simpleness of search has trained users to simply type in boxes and expect a result. Most users won’t bother to drill down through a category list.
When I list stuff on eBay (where, again, I have a financial incentive to categorize things), I just accept whatever category their search box tells me to put something in.
I agree that the classifications others have created can be effective in training algorithms. I’ve long advocated using datasets like The New York Times archives (classified by professional librarians) as training sets for search engines.
The challenge is that the data set has to be accurate, comprehensive and consistent. Combine the large volume of data that is being created today with the pollution efforts of spammers and this gets harder and harder.
Thanks for the pointer on the ODP citations in patent applications. Interesting stuff.
> “My post was in reference to the navigation value of classification systems. The effectiveness and simpleness of search has trained users to simply type in boxes and expect a result.”
That´s very true, of course. And a challenge not only for classification systems, but for search engines too, if they want to tempt the searchers to use any related search, tip or advanced search feature. Google has conditioned people to type keywords into a box and expect miracles, and now everybody has to live up to the expectation 😉
> “Most users won’t bother to drill down through a category list.”
Hm… now here’s a heretic question: Why should they have to “drill” from category to category? Why does directory design not encourage other types of browsing?
Imagine there’d be an alternative view, which displays the taxonomy of a classification system like a map: you could go to bird’s perspective and see all categories on the next three, five, or more levels before and behind you at once, or zoom in on a specific category…?