askme sample page improvements

December 15, 2004 9:01 AM

Recently, Matt put up a sample page for the new AskMe Ask A Question page. Included in the sample are sections for questioner selected categories and keywords. The cataloger in me sees both the use and the danger in this, so... [+]
posted by robocop is bleeding to feature requests at 9:01 AM (43 comments total)

The idea is to make AskMe more searchable and therefore more accessible to users. This is A Good Thing, but I'm concerned that without some level of quality control, this feature will become unhelpful, or worse, counter productive. I raised some of these concerns a few threads back, but the thread sorta changed direction (or maybe I was in the wrong place).

First, the subject categories.
The categories listed on the sample Ask page are: clothing, beauty, & fashion; computers & internet; education; food & drink; grab bag; health; home & garden; human relations; law & government; media & arts; pets & animals; religion & philosophy; science & nature; shopping; society & culture; sports, hobbies, & recreation; technology; travel & transportation; work & money; writing & language.

Do these categories adequately reflect the sort of questions that we see on AskMe? While they are good categories for a general knowledge base, I don't think they reflect the proportion of question subject normally asked in AskMe. To check this out, I picked a random day (methodology: dart at calendar) to catalog based on the 20 subjects: November 23rd 2004.

Of the 40 questions asked that day, it came as no surprise that the most asked subject was Computers & Internet (work & money - 5; travel, clothing, and shopping tied at 3). While quickly cataloging, I was struck by some of the questions that had arguments for being in multiple categories (web archives of comics - media or internet? shopping at Ikea - home or shopping?) and some of the variations within single categories (work & money - there was a retirement question, one on non-degree requiring jobs, and one on quoting charges for website design). The variations were enough that I am wiling to be that if another person did the cataloging, they'd come up with different results (I put all the New York vs New Jersey in society & culture, for example, while others could stick it in human relations or even grab bag).

So what's all this mean? In my mind, it means that like the three bears' sleeping arrangements, some of the categories are to big, some just right, and maybe one or two too small. It also means that there will be some variation when users select their own categories.

Like I mentioned previously, it may be a good idea to break out some of the subject categories. After looking at Nov. 23rd, I'd want "computers & internet" broken down (Computers - hardware and computers - software, maybe?) for sure. I'd also want a "regional" category for questions about specific locations outside of "travel & transportation".

What other categories would be useful? At what point are there too many of them and everyone just ends up entering "grab bag" for fear of a long, nebbish Mefibrarian post like this one?

Of course, to help break down all these subject categories, Matt has included keywords. That's the second kettle of fish.

First, should keywords be required? If so, how many? If you require all three, some people may be tempted to just enter random words or phrases that have little to do with the question in order to post faster. If you require none, then people may skip the step entirely.

Second, what quality control is attached to keywords? People post to MetaTalk when a rogue question pops up on AskMe. Would we really want them doing that to quibble about a keyword's meaning or point out that every fifth keyword is "quonsar"?

Third, can there be any form of authority control to keywords? That is, can there be a database that informs the AskMe keyword search that "tv" is just as good as "television"? Looking at the sample AskMe page the three keywords are ipod, audio, and auto. So what happens if I search for ipod, stereo, and car?

Fourth, how useful are keywords that appear in the text of the question itself? Would the search feature bring up both keywords and text?

I have no idea how to answer these questions from a coding standpoint, but from a searcher's standpoint, I'd want at least one keyword to be required, for there to be some form of quality control to make sure that word is relevant, for there to be a thesaurus linking words like "car" and "auto", and for the keyword search function to also include "Your keyword appeared in the text of _____ other questions. Click here to view them."

Okay, this is not only my first MetaTalk post, but it's really long. I'm interested in what y'all have to say, though. I could be fixating on nothing.
posted by robocop is bleeding at 9:02 AM on December 15, 2004 [1 favorite has favorites]

I think this is awesome. I'm sold.
posted by Quartermass at 9:10 AM on December 15, 2004

See, the thing that jumped out to me on the new page was the word "jist", which should be spelled "gist", but that's just the pedant in me.

More on topic: I used to work for a company that did asset management and we could never get user-defined, free-form categorization to work. Most people aren't taxonomists and when you let them all come up with ways of classifying their own work, you end up with a big mess. There needs to be some editorial control, if not on the categorization itself, then on the list of categories.

Beyond that, people will use freeform keywords for wholly inappropriate ends. Look at the "Title" field for FPPs. The intention was to have a place to summarize a post so it would look nice in RSS readers, but everyone uses it as a place to put a joke. In the end, the usability of it is gone.

My proposed solution?

1) Create more nuanced categories (robocop's "Computers > Hardware" and "Computers > Software" example being the right idea)

2) Come up with a methodology for users to propose new categories that mathowie could add to the system. Or else the Metatalk homepage will fill up with new category requests and crowd out all the pleas for n00b tolerance, which I totally love, by the way.

3) Let people select multiple categories. Ban the people who select EVERY category.

4) Get rid of keywords altogether.
posted by turaho at 9:29 AM on December 15, 2004

i think it was Stanchin who proposed a category of "my fucking cat". That would be a good one.
posted by kev23f at 9:29 AM on December 15, 2004

I'll revisit these points after the new page has been up for a month, and I'll tell you why.

A cataloging friend scoured a year of archives to come up with the categories to cover every question asked, but yeah they could use some tweaking, which is something I'll do after it runs for a while. If there are too many grab bag questions, those will become new categories. If a category never gets a single post, maybe it'll go away and end up as rarely used grab bag question.

On the keywords, everyone said Flickr and del.icio.us were wasting their time with keyword tags because without standards, they would be useless, but you know what? They're pretty useful even with the "dirty" nature of the data collection. Again, something else to look at down the line, after ad-hoc standards emerge.

I don't want to overanalyze something before it has even been released in the wild. First it has to come out, then we should let it run for a while to get some data to see if these worries are warranted.
posted by mathowie at 9:33 AM on December 15, 2004

"...gonna stab the next person who proposes coming up with a tag scheme to be used by everyone..."

OK, so I disagree with the above re:categorization. Tags should be free form, and we should have more than three. - in a perfect world we should have many tags. But I can see why only having three would reduce the complexity of implementation many magnitudes. Also, at some point, too many categories becomes impossible to use - do you want Computer, Computer:Mac, Computer:OSX, Computer:Win3......? You could have an infinite amount of categories and someone would be complaining about it. I would prefer to see only del.icio.us-style tagging, because hierarchical categories are dead. A bit overboard yes, but with a userbase of over 20,000 creating uniform standards is impossible. MeFi users are not books to be assigned meaning via Dewey or LoC or some other hierarchical top-down system du jour. Long Live the Folksonomy! (self-link) Matt's combination of tags and broad categories is about the only useful compromise out there (for now).

As for mandating a "tag police," people will self-police their tags. Look at Flickr or del.icio.us. As for the coding/implementation of tags re:combining tags (ie computer+mac) it is quite hard to get it right, just ask Joshua Schachter. Also, you can google search AskMe, which makes things really easy. Why create a huge server and administration load when you can implement simple, broad categories plus tags. Finally, I'm of the "let's wait and see how this works" and slowly roll things out. I suppose this fits with how much time Matt has to implement things.
posted by plemeljr at 9:33 AM on December 15, 2004

4) Get rid of keywords altogether.

Ha! You data standards nerds crack me up. I swear to you that they will be 1000 times more useful than categories ever will. Trust me on this one, and watch.

Keywords are gold.
posted by mathowie at 9:34 AM on December 15, 2004 [1 favorite has favorites]

My proposal would be the opposite of turaho's: get rid of categories altogether, and use keywords just like del.icio.us and flickr use tags. On preview: what plemljr and mathowie said.

I have one additional concern with tags here, however. At del.icio.us and flickr, tags are always edittable. I would guess that tags attached to an AskMe question here would no be. That gives us a strong incentive to make sure that querists get the tags correct the first time around.
posted by monju_bosatsu at 9:36 AM on December 15, 2004

I thought that the keywords and the categories would work together, so that although you can only pick one category, you can use the keywords to suggest other areas under which the question falls. I don't think any arrangement of categories will ever be immune to questions falling either between the cracks or into multiple categories.

(on preview: this thread grew fast)
posted by kenko at 9:37 AM on December 15, 2004

1) Create more nuanced categories (robocop's "Computers > Hardware" and "Computers > Software" example being the right idea)

There is plenty of precedence for this, on Google Answers, on the DMOZ directory, on Yahoo, and I think the consensus is that while librarians are stoked with these taxonomies, people don't much care for them. They're tedious to enter new data in, since you're always worried you're in the wrong segment of the deep, deep category system, and the maintainer has to constantly tweak the taxonomy to represent any concept on earth, and that has to fit everyone's mental model.

I didn't even want to have categories, but a good librarian friend said we should try simple flat ones to see how it goes, and I'm still sure the keyword tags are going to steal the show and show their simple yet powerful worth.
posted by mathowie at 9:38 AM on December 15, 2004

Keywords are gold.

Dude, keywords are categories.
posted by kenko at 9:39 AM on December 15, 2004

This may be a coding nightmare, but what if it wasn't the poster who selected the catagory, but the rest of the users? Perhaps the catagory it's filed under could be what the majority of users selected?

So, when reading a question on AskMe, a user could have the option of picking a catagory that they think the question should be filed under. That way, those of us who aren't organized enough don't have to select a catagory, but the post will eventually be filed properly.

Personally, I'd just use a default catagory for all my questions. I hate having to pigeonhole things, though I can understand the use of it.
posted by bondcliff at 9:43 AM on December 15, 2004

I would guess that tags attached to an AskMe question here would no be.

editability could easily be there.
posted by mathowie at 9:47 AM on December 15, 2004

Matt, a suggestion: let the keywords be selected from a list, which would contain the keywords already used for a certain category, ranked by popularity. (They still should be editable, if the right keyword is not in the database. Use max top 20-40 keywords, not more for each category. The keywords should be linked to a category, a global list of keywords might not be a good idea.)

From time to time you would have to eliminate some keywords manually (e.g. in ‘car’ vs ‘auto’ keep only ‘car’) and assign the synonym keyword to, for example, ‘auto’ posts (only if there is a simple way to do this, otherwise is not worth it). After few month the system should be self sustaining.
posted by MzB at 9:53 AM on December 15, 2004

A character in Vonnegut's Cat's Cradle tells the protagonist "Never let authors write their own indexes." I think there's some truth to that, because what's important to the person asking the question may not be as important to the user looking for similar questions.

I really do think the keywords channel will fill with noise pretty quickly, Matt. But you're right... let it out in the wild for a little bit. I'm constantly amazed how AskMe stays mostly noise-free (up for debate, I know) so maybe this will work out, too.
posted by turaho at 10:04 AM on December 15, 2004

Third, can there be any form of authority control to keywords? That is, can there be a database that informs the AskMe keyword search that "tv" is just as good as "television"?

I have faith in the categories. However, this issue is my only concern. Will I need to know to search for weed, pot and marijuana to make sure I'm doing a good search? Is there some useful way to alert people to what the current categories are... maybe even a pop up window. MzBs idea is really a good one, though I think I'd allow user editable keywords first and see how that shakes out, maybe lock them up after the keyword database is mostly seeded and just have one user editable category.

Severe penalties for anyone abusing keyword functions by the MeFi librarian posse & mathowie. Some good guidelines would be key to answer questions like

- do you have to add keywords that are also in the text, for example? [I vote no, keywords are to add retrieval handles not to make them redundant, then again, I could see a very good argument for a keyword-only search getting more relevant results than a full text search so maybe that would be useful...]
- for regional questions, do you need city AND state AND country? Should a search for Arizona get Tuscon questions and vice versa?

I vote case-insensitive keywords, I assume this is a foregone conclusion? I don't want to have to mess with iPod or ipod or Ipod.

What about stemming for the search engine? So a search for libra would get library, libraries and librarian?

I would change "human relations" to "love, sex & relationships" but maybe it's also supposed to be there for fights with your roommate? It will be fun to see how this all winds up.
posted by jessamyn at 10:09 AM on December 15, 2004

"Keywords are gold."

Dude, I know keywords are all the rage now, but keep in mind that you already experimented very lightly with a visible and reasonably important piece of metadata: MetaFilter post page titles. And, in case you haven't noticed, most people fuck those up pretty badly.
posted by majick at 10:26 AM on December 15, 2004

Top-down taxonomies are always hard: want a hair-pulling, vein-popping argument? Get a bunch of taxonomists together and ask them to categorize a new kind of fungus.

Flickr/del.icio.us tagging works surprisingly well. I think #1 has the right instinct here---with one suggestion: How about showing a list of the 25 or so most popular tags on the right when posting a question, sort of suggestions for tags, if you will.
posted by bonehead at 10:27 AM on December 15, 2004

you already experimented very lightly with a visible and reasonably important piece of metadata: MetaFilter post page titles. And, in case you haven't noticed, most people fuck those up pretty badly.

This point was made above, too, and I finally figured out why it doesn't sit right with me. There is a long tradition of creative titling for all kinds of pieces - fiction, newspaper articles, you name it. Geeks think "titles are metadata!" but a lot of normal people think that titles are part of a narrative whole. I don't think "keywords" evokes the same level of desire to display cleverness.

Hope I didn't jinx it, you smartasses.
posted by caitlinb at 10:34 AM on December 15, 2004

Hi all! I was hoping there'd be some response after I got back from lunch and lo, I am not disappointed.

I'm willing (like it even needs my stamp of approval, eh?) to wait out the categories and fine tune later, if needed. My concerns could be just the product of too much caffine. I'll be interested to see how useful some of the categories are, but that could be fixed on the keyword end (provided the whole car/auto thing works out - MzB's idea is pretty neat).

One other AskMe pony I'd like to see would be for there to be some way for the questioner to show, in a searchable form, what sort of answers they are looking for. Are they looking for product recommendations? The experience of other users? Technical advice? Opinions? What would others do in a similar situation? etc.

Even though two questions subject categories and keywords can match, what the two authors want from their questions can be completely different. For example: Both the questions "Any advice on starting a rock band in NYC" and "What's the best concert venue in NYC?" could have the subject "media & arts" with keywords "New York", "entertainment", and "music". The former question is looking for advice, while the second is looking for an opinion.

I was able to explain this a bit better on my drunken napkin awhile back. Does anyone get what I'm talking about, or am I rambling again?
posted by robocop is bleeding at 10:44 AM on December 15, 2004

most people fuck those up pretty badly.

that's an assessment, not a fact.
posted by quonsar at 11:11 AM on December 15, 2004

P'raps the solution to keywords and catagories is to have them assigned not by the poster but by any of the great number of librarians prowling Metafilter?

More when I'm a bit more awake.
posted by stet at 11:21 AM on December 15, 2004

I think the preselected keywords and/or specific categories would be extremely helpful. It's nice to go to a site looking for info and be able to click on Computers->Software->Games->Adventure. As someone mentioned, keywords are categories when you get to this level. Also, doing things this way doesn't mean giving up anything, just adding.

On preview, stet's suggestion has potential. Maybe we need volunteer archivists. I'm sure there are many established users who would be happy to help.
posted by spaghetti at 11:34 AM on December 15, 2004

you already experimented very lightly with a visible and reasonably important piece of metadata: MetaFilter post page titles. And, in case you haven't noticed, most people fuck those up pretty badly.

Titles are only used in RSS and at the top of the browser where seldom look, and seldom to people take them seriously. Why? Because there is little utility to them.

With keywords, certainly someone will fuck around every once in a while, but the keywords are going to become a very active navigation tool. Every keyword will be a hyperlink to every other post ever, with that same keyword. I envision a new sidebar on the front of ask metafilter with the last x keywords and top x keywords. They'll be infinitely more useful than page titles, hence I suspect the tomfoolery to be minimal, since good keywords will help everyone find their answers easier.
posted by mathowie at 11:39 AM on December 15, 2004

The thing about the MeFi titles sucking is probably mostly because the titles serve basically no purpose. So horsing around with them doesn't seem like you're doing the site a disservice.

I think there's more incentive to play along with keywords. It helps the site, gets your question read by the right people, helps you find your old questions, etc.

(on preview) Matt and I are the same person.
posted by frenetic at 11:41 AM on December 15, 2004

Matt, any timeframe on any of this? With the understanding -- nay, the certainty -- that it would be loose?
posted by stupidsexyFlanders at 11:51 AM on December 15, 2004

Great developments, and I'm of the release an iteration and wait and see camp.

On robocop is bleeding's specific idea here, I think even something as simple for answers sought as: Facts, Opinions, or Both (facts and opinions) could be effective.

However, I think most folks on askme are already looking for facts and/or opinions in the form of advice so long as the response is relevant to their question, so ultimately is a pony like this worth its resource drain to implement? Probably not, but it's still a nice thought.
posted by safetyfork at 11:58 AM on December 15, 2004

The best AskMe questions not only defy categorization, they're not something you'd search for without already knowing they've been asked. (I guess there could be better examples) I think it was the lack of structure, combined with community feedback, that nurtured these. I fear AskMe is that special jewel that will dissappear the minute we try to reach out and grasp it.
posted by klarck at 12:07 PM on December 15, 2004

safetyfork, what if the "Intent Information" was a series of checkboxes? So if I'm looking for help with virus software for a PC, I'd check product recommendation (What should I use?), experience (How have you dealt with viruses?), and advice (What would you do in my situation?).

Now, it'd be really cool if we could catalog the answers along the same lines, so in theory someone could search for "Give me all product recommendations in the health category with keyword acne" or "What are people's opnions about subject: media & arts and keywords: indie rock".

But like you said, the work vs. result ratio may be wanting.
posted by robocop is bleeding at 12:12 PM on December 15, 2004

Mmm...I think it would be cool (check boxes and all), but probably a little too robust and costly to go under-utilized which it seems like it might.
posted by safetyfork at 12:31 PM on December 15, 2004

GO LIBRARIANS GO! YAY!
posted by fishfucker at 12:35 PM on December 15, 2004

I find that too much structure, too early can kill any project, and suspect a lot of these ideas presented here are adding too much complexity, in order to tease out a little info further down the line.

I'm going to go with simple categories, with keywords and more attention paid to keywords and see how things go from there. The one comment I get most often about Ask MetaFilter is how drop-dead simple it is, and how amazingly well it works for being so simple.
posted by mathowie at 12:42 PM on December 15, 2004

Keywords are gold.

Yeah, I know. That is why I end up spending so much time at the reference desk helping people who just spent 20 min. (or 2 hours, or whatever) trying to find something on google, and who failed.

Although, considering what I get paid, I think keyword-based systems are more brass.
posted by QIbHom at 1:32 PM on December 15, 2004

As far as multiple versions of the same keyword goes (car, auto, automobile, Car, autmobile, horseless carriage) it is pretty easy to fix that retroactively with scripts. Initial building would take a bit of time but after a couple dozen iterations I'd bet you'd be down to one or two items to add to the script a month. I've done this kind of clean up for addresses and although a simpler problem (there is a single known right answer) it's basically the same concept. My script to correct street names in a town needed only minor tweaking once I'd fed it enough projects.
posted by Mitheral at 1:47 PM on December 15, 2004

Not a comment on the categories specifically, but a request to Matt on the new Ask-A-Question page: can you change the language for the FPP and make it a multi-line text box? FPPs like "What is this book?" and "How do I fix Firefox?" really grate. There's a reasonable cutoff point at which details of a question should go inside, but "just a few words" is not it, IMO. A few sentences is more like it. Frankly, I think the entire example you give for the "More Inside" on the page ("I just bought a new Jetta, and I have a third generation 10Gb iPod that I'd like to use. I've tried out the iTrip but the FM stations are too strong in my area. What are my other options?") is not too long to be on the FP. Please keep in mind that not all of us have time to (or want to) go into the thread for every question, so the FPP needs to be descriptive enough for people to decide whether the question interests them/it's something they might be able to answer.
posted by DevilsAdvocate at 4:44 PM on December 15, 2004

I've typed far too many words on this subject in previous threads, but I'm very pleased to see that Matt has drunk the folksonomy-keywords koolaid and is going ahead with a full implementation of what I was hounding him for months back.

This'll be cool.

Now all we need is the same keywording functionality throughout the *.metafilter.com sites! ;-)
posted by stavrosthewonderchicken at 4:57 PM on December 15, 2004

keep in mind that you already experimented very lightly with a visible and reasonably important piece of metadata: MetaFilter post page titles. And, in case you haven't noticed, most people fuck those up pretty badly.

Maybe I'm an idiot, but I had no idea those were for RSS readers. Not using RSS myself any more, it never occurred to me. I visit MetaFilter nearly every day, but maybe I was distracted that day. Point is, maybe it's just a knowledge/publicity issue? I don't know.
posted by gd779 at 7:20 PM on December 15, 2004

I used an RSS reader to read mefi for a while, but the shitty page titles were just too much to handle. I could never figure out what the story was because people always screwed it up. Thats why I always post useful page titles.
posted by puke & cry at 9:50 PM on December 15, 2004

From a user/viewer perspective I would suggest that at least one of the keywords validates against the title, short question or extended question for use. This should work as an easy threshold against abuse as well as migration to higher 'categories'.

Having worked with a similar set up as you propose one of the 'down the road' issues I encountered is that keywords slowly became a free for all category fest and may lean themselves to bending an otherwise reasonable structure. The meta* sites are recursive in nature and often focus on internal data and slang so keeping the core idea of at least on keyword validation will help to make the noise more navigable.

I may be way off, and free for all does permutated into the real nature of things, but a single worked would be a good guiding practice.
posted by dasibiter at 10:05 PM on December 15, 2004

Both categories are tedious and PITA and keywords don't scale.

The solution?

Every user who posts a question to AskMe should have to link to three previous questions that are 'related'. All cat posts should link to previous cat posts, all book posts should link to previous book posts, all 'living on a budget' posts and so on. This also ensures users have done due dilligence.

Hyperlinks are the most powerful information organizing force the world has never known. You'll see. Someday you'll all see...
posted by nixerman at 12:58 AM on December 16, 2004

What if the keyword fields work a la Google Suggest, so if you start typing "pet" it gives you a list of already-used categories, and the number of posts with that category? That might help prevent duplicates.

So "pet" might match:

"pets" (50 posts)
"petfood" (6 posts)
"peterpan" (1 post)
posted by gramcracker at 7:42 AM on December 16, 2004

If keywords are gold, bondcliiff's idea must be platinum. I wouldn't just go with the most often assigned keyword, however; store all keywords selected. Then, if someone searches on "cat", e.g., posts could be displayed in descending order of the number (or percentage) of times "cat" was entered as a keyword. If someone, perhaps even the original poster, while under the influence of cough medicine or paint thinner chooses something ridiculous like "marmoset" as the keyword for the post it'll get swamped by more appropriate keywords. In addition, people who assign peripherally related keywords, like "pet" e.g., will stilll be contributing to the "searchability" of the keywords, just not as strongly as those assigning the most common keyword. So people searching on "cat" or "pet" will eventually arrive at the post. So would people searching for "marmoset" but nobody would need to search on "cat", "pet", and "marmoset" to find the post in question.
posted by TimeFactor at 6:22 PM on December 16, 2004

do you have to add keywords that are also in the text, for example? [I vote no, keywords are to add retrieval handles not to make them redundant, then again, I could see a very good argument for a keyword-only search getting more relevant results than a full text search so maybe that would be useful...]

Jessamyn--this broke my brain, can you 'splain? To me, it's obvious that most keywords will be in the post. If my question is "How do I hook my iPod and my toaster together so I can store a thousand varieties of strawberry jam?", I'm going to keyword "ipod" and "toaster", so when folks click on the keyword "ipod" somewhere else, up comes my post. What would a non-redundant keyword be like?
posted by frykitty at 9:00 AM on December 17, 2004