Imagining Better Icebreakers

K CUPIDZ LATEST FEACHUR

OkCupid boffin winteraz84, in his recent journal post entitled The Inside Scoop on Icebreakersprovides a detailed explanation of what this new feature was imagined to be, how it was built, what problems it was meant to address, what problems arose in attempting to make it work, and the challenge of getting anyone to read any explanation of how anything works anyway (i.e., "the 7 second internet").

WUT WUZ WRONG?

To summarize his deeply enlightening write-up, Icebreakers were designed to address three (and then, after the fact, four) problems (which I shall give a different spin while still trying to capture winteraz84's intended spirit):

1. Pressure:  Unlike casually meeting someone and feeling them out in conversation, messaging someone on a dating site is an one-shot, all-or-nothing endeavor.  Simply put, the stakes are higher, and this impacts on the likelihood that one will even take the gamble.
2. Silence:  Many people (especially men, women still expecting to be pursued, not to pursue) go for long stretches (the bar used by OkC is a week, but we all know it can be much longer) without getting any messages.  Quite frankly, that's no fun.
3. Absence:  In light of #1, above, who wants to craft a carefully tailored e-mail to someone who hasn't been on the site for months?  Like they're really going to log in because yours truly wrote them, right?  Well, winteraz84 suggests that ~1/10 of those messages do get replied to.
4. Crassness:  Eliminate the incentive of eye candy, and you might potentially cut down on the volume of first time e-mails that verge on sexual harassment. This one seems a bit more dubious as a likely benefit, but implicit is the expectation that all such hopeless pervs need is a little help recognizing that they, themselves, have interests other than how hot someone is.

TEH AMAZIN IDEA

The solution hit upon by the OkC boffins was "blind one-to-many messaging".  Rather thanblind, which sounds suspiciously like a recipe for spam (which is one of the things the feature is intended to combat), it might better be understood as hand-held while blindfolded.  The idea being that such a system would be:

1. Fun:  low pressure; risk more akin to spin the bottle than Russian roulette; restores ambiguity to interactions.  All a formula for less stressing and more messaging.
2. Mysterious:  boffin wisdom is that chicks dig this; counters swoon-generated second-guessing.  Like a Secret Santa (or, for us Pennsylvania natives, Pollyanna) gift, a low investment way to increase total messages exchanged overall.
3. Polyvalent:  two messages, no three messages, no X messages for the price of one, folks! It's the days of carbon-copy all over again; don't fill it out in triplicate, just press (i.e., think) real hard to make sure it comes through on all copies.
4. Guided:  when "tl;dr" is default behavior, shield the user from text walls; give them space, draw their attention to things they like, and encourage them to play (going back to #1).

UH OH, IT NOT WERKIN!

However, in the course of development and initial deployment of the new system, problems were identified:

1. Temporal:  original prompt "What are you doing this weekend?" presented issues of latency (message might be read after the fact) and co-temporality (being on a date with three people at the exact same time might make for a funny sitcom, but not so cool in real life).
2. Brevity:  or, to put it more rudely, many folks don't fill the sh*t out; kinda makes for poor data mining opportunities when looking for interests.
3. Inanity:  like, I [[really]] don't need [[to]] explain this one, we've all seen or [[done it]] (so much for Sir Tim Berners-Lee's vision of a semantic web).
4. Stupidity:  this is referring to the limitations of Staff Robot, not us users--COMPUTERS JUS DOAN GIT WUT US HUMANZ IZ TALKIN BOUT.
5. Negation:  related to #4, but a distinct issue in natural language processing: if you've ever written to technical support (for any company, take your pick) and said I have a problem, and it is not XYZ, you have gotten the obligatory automated reply explaining to you how to fix XYZ.
6. Buttons:  the original idea had been to address problem #5 by giving the user an option to clarify their intent when they said "I'm allergic to cats."  However, the KISS rule (Keep It Simple, Stupid) won the day, at least initially.
7. Transparency:  tell people they are blindly messaging a group, they get stupid; don't tell them, they get mad. In the end, a thick, light-proof blindfold was deemed preferable. 'Nuff said on this point.
8. Balance: want every broken ice cube to sparkle with potential, only offer them to people when they have more things in common than they have fingers--with not one, but a dozen other people.  Want to ensure that most everyone never, ever, in a million gazillion years, get to use the feature, do the same.

Having provided that helpful summary (more for my own edification than out of any expectation that anyone has read this far), here's a scenario I would encourage winteraz84 and his fellow boffin brethren to consider:

BRACKETS, WE DOAN NED NO STINKIN BRACKETS

First off, among the problems identified above, #3 would, I suggest, have several origins.   GenericMale identifies the first of these in his comments to winteraz84's journal post.

• Oooh Pretty:  absent CSS or some other beautifying treatment, users have creatively re-purposed them those double brackets for the purposes of emphasis, highlighting, marking headers or sections, and/or satisfying the bad habits they developed on MySpace.
• No Comprende:  many users don't understand what the brackets are for, and only use them (and often don't even use them successfully--oh, I was supposed to hit that funny key *twice*?) because Staff Robot keeps nagging them to.  Out of laziness and/or spite, they drop the brackets wherever.
• Big Brother:  many users understand perfectly well what the brackets are for, and quite frankly it creeps them out. Maybe they've answered match questions (and maybe they haven't), but they still have a vague notion that if someone with more power than they is listening for keywords, their best strategy is to create more noise than signal.
• Zero Confidence:  some users, those who have been here a long while, understand what the brackets are for, and really wish they did work, but they've been around the OkCupid block a few times, and they just don't see it happening, so why bother pretending.

Combine this with problems #2, #4, and #5:  not everyone writes a paragraph, let alone completes all the boxy-thingies; even if they did, Staff Robot (like all robots--it's not that we don't love our Staffy) does a very good impression of a functional illiterate; and the one thing you'd think a robot should be good at, boolean logic, only really works with numbers.  What you get is a very powerful idea, limited by its own input stream.

It's not so much GIGO (Garbage In, Garbage Out) as no-go, for without (valid) interests to slush around in the Icee maker, you don't get to fill your cone.

OH NOEZ, 2 LOTZ DA WERDZ!

Not everyone is equally good at everything.  That's part of what makes meeting new people so interesting.  You never know what unique knowledge, skill or ability they might have that you don't, or vice versa.  By way of examples:

• some people never learn to ride a bike;
• some, round these parts (New York City, thank you for asking), never learn to drive a car;
• some never quite get the hang of all the bells and whistles on their car stereo; and 
• some sure as hell are at a loss when they have to find their way around a computer screen crammed full of clutter...

...although they may very well have learned how to ignore anything that looks like it might be:

• a graphic (probably someone trying to sell me something I don't want), or
• lots of text (probably someone trying to tell me something I don't care about). 

This is why the decision was initially made to not include an opt-out button ("No, Staff Robot, I really meant it when I said that I HATE THE METS!") within the Icebreaker messaging interface.  I would argue in favor of this original decision, for reasons I'll go into in the next to the second to last section.

NOTHIN 2 C HER, MOOV ALONG

Similarly, not everyone is naturally verbose when it comes to the written word:

• Too Hard:  some people don't enjoy writing, find it deeply challenging, even.  They could be mathematical geniuses or the sort of person you want on hand for all your DIY projects, just don't expect them to pick up a pen or peck at the keyboard if their grade point average or paycheck doesn't depend on it.
• Tough Subject:  some are comfortable with writing, but describing themselves isn't what they do well.  Perhaps they aren't the deeply introspective sort.  Maybe they are better at writing technical manuals than marketing brochures.  They may worry they'll give a wrong impression or would rather people learn about that World of Warcraft addiction only after the first date.  Perhaps they were brought up to believe that talking about yourself is the best way to make a bad impression.
• Why Bother:  many are convinced that no one is going to read what they're going to write, and choose not to waste their time.  Keenly aware of the 7 second rule, they act accordingly.  Or, just as keenly aware that pictures speak a thousand words (and that people often stop thinking once they see a picture), they likewise don't waste their time at the keyboard.

On the flip side, much of the content that appears in the more densely written profiles is, face it, nothing but lists.  What's my favorite music?  Here, wade your way through this 20 line text wall, I dare you.

HOW WUD U LIEK UR ICE BROKD?

This brings us to my suggested improvement to the Icebreaker engine:  just as you encourage users to improve their matches, allow users to refine their ice.  I would suggest this as taking up the slack of what both brackets and the "My favorite books, movies, music, and food" questions do so poorly.

The use case might be imagined as follows:

1. User clicks on the Icebreaker tab.
2. If potential icebreaker candidates have been identified, display the current messaging screen, inviting the user to craft something to be sent to their mystery match(es).
3. User either sends a message or declines to do so, or if no Icebreaker match today, skip to step #4.
4. If the user has ungraded bracketed terms, present one of those terms and prompt "Please help Staff Robot to find you better mystery matches, but telling us how much you like ______."  Followed by some sort of enumerated scale (e.g., a Likert scale), and also a "WTF is '______'?" choice.
5. If the user asks why _____ is being presented to them, use it as an opportunity to provide a short explanation of how brackets work, and give them the option to remove the brackets from that term and/or skip grading that term.
6. Repeat step #4 until all non-skipped bracketed terms are graded by the user, or user decides they've done enough of this exercise for today.
7. Present much the same prompt as in step #3, only this time using terms found in user's profile that:  a) tend to be liked or disliked by other users; b) are liked by potential icebreaker matches for this user; and c) fit neatly into a taxonomic category (will explain this below).  Let "neutral" option double as "nonsense", or include a separate option.
8. Repeat step #7 until all terms meeting criteria 7(a), (b), and (c) have been graded, or the user decides they've done enough of this exercise for today.
9. Based on the interests graded in prior steps, draw some inferences as to other things this user might be interested in, based on data for other users with interests similar to theirs.
10. Prompt "Based on the interests in your profile, Staff Robot wonders how much you like _______."
11. Repeat step #10 until user decides they've done enough.

Y EVR WUD U DO SUCH TING?

The benefits of a system as described above would be multiple:

• Clickity Mousey:  Users don't like clutter, they don't like text walls, but they don't seem to mind ticking a few choices and moving on.  The success of the OkCupid matching system proves this.  What about users who don't bother to answer matching questions, you ask?  Icebreaker wouldn't work for them, anyhow (as it takes match score into account for potential mystery messages.)
• No Clutter Messaging:  The user might be just a bit perturbed the first time (step #2) they see that Staff Robot has assumed they really like clowns when what they said in their profile is that they are deathly afraid of clowns, but when they click the "No Thanks" (in step #3) on the messaging screen, and the next thing they see (in step #4) is "Please help Staff Robot to find you better mystery matches, by telling us how much you like clowns." they'll perk up and admit that Staff Robot really is trying to help, after all.
• Richer Data:  Not only will an enumerated scale (step #4) allow you to eliminate those pesky negatives lurking in the natural language, they will also allow you to identify items that the person is quite neutral on, but which they mentioned just for rhetorical effect.  ("I don't care about __sports__, but I won't mind cuddling with you while you watch the game.")
• Superlative Data:  What's more (still step #4), knowing that someone likes Morrissey way more than Billy Joel, even though they enjoy both, could come in handy for compelling mystery matching.  ("...she's between the ages of 23 and 35, likes Billy Joel and tennis, and loves Morrissey.")  Behind the scenes it can provide extra weight in scoring potential mystery matches.
• Bracket Education:  Here's your 7 second opportunity (step #5) to Stop the Inanity!  What better time to explain to a user why "[[two]]" or "[[why I even bother with this site]]" are not in the least bit helpful, then when you are explicitly asking the user to help you help them?  Now you've got their attention!  For a few seconds, at least.
• Collaborative Fill:  By letting user responses guide Staff Robot's choices for potential terms of interest (step #7), words and phrases that didn't make it into the system fill-word dictionary, or that are not otherwise recognizable as non-germane parts of speech, such as partial names or netspeak, can be accounted for.  If a ninety-nine of the last hundred users said "lolz" is not an interest, user 101 doesn't need to be asked how much it interests them.  (On the other hand, if they really want it marked as an interest, that's what brackets are for.)
• Mysterious Guidance:  This is pretty obvious, given the whole purpose of Icebreakers, but bears being stated explicitly just the same: the first bare terms from a user's profile (step #7) that should be asked about are those that, if the user identifies them as interests, would push potential three-interests in common matches into the four-interests in common bucket, etc.
• Pull Down Listing Walls:  The best data to use for this (step #7) would be that which can be checked against a public data store for a specific category of interests, such as musical artists or movies.  The "My favorite..." question could be replaced with something far more powerful features.  Imagine visiting a profile and seeing a message:  "You have 10 favorite musical artists in common with this user! Click to find out more!"  
• Overcome Brevity:  They may not have written much in their profile, but that doesn't mean you can't figure out what their interests are.  So long as they've written *something* in their profile that is recognizable as an interest, Staff Robot should at least be able to make some educated guesses (step #9), and then ask them to confirm (step #10).  Who knows, once Staff Robot suggests that their addiction to Mystery Science Theater 3000 as a teenager might actually be of interest to someone, they could be inspired to go back and actually write about that in their profile!

UM, R U FINISHD NAO?

The potential feature possibilities from categorical terms don't stop here, either.  Imagine integrating with other social networking sites like last.fm, blip.fm, goodreads.com, or even the RSS feed of recently watched shows on hulu.com.  Mining such third-party data could greatly expand the potential terms for bringing together mystery matches, while saving the world from text walls and lists that never end.

Or, imagine coming to someone's profile, looking at their favorite movies, and being given the option to indicate which of those movies you had forgotten to list in your own profile, but really like.  ("How could I have forgotten Being John Malkovich?!  I love that film!")

All in all, the Icebreakers feature, now that its purpose has been communicated, looks like a smart move on the part of OkCupid.  What is proposed here is an expansion on that original idea, with consequences for improving not only the Icebreakers experience, but potentially also the form that profiles take and the ways in which users share their favorite interests.