We might be missing the point here by focusing on romantic relationships.
The fact that people make fewer posts on Facebook as they become closer might mean something, in general, about the kinds of relationships that Facebook sustains.
Facebook is the modern equivalent of a Christmas card...It's a way of maintaining surface-level contact with many people: letting them know that you care...but you don't necessarily care enough to catch up through a more direct and personal medium.
Communicating with your closest friends or your partner mainly through a Christmas card would be silly, in my opinion; we tend to choose less distant methods of communication. So, it's not surprising that people would exchange fewer posts with close friends, as well as romantic partners, as they get to know them.
Those are just my two cents, speaking from experience and observation. I'm sure HN has a range of opinions about this, and I'm interested in hearing other people's perspectives.
I think the sociological term you are looking for is "weak ties" [1] - Facebook (minus Messenger) appears to be a platform for weak ties more than strong. There are social, strategic, financial, and political benefits to maintaining a network of weak ties [2].
I post things to the walls of close friends fairly often, because I know them well enough to know what they'll find interesting. Also, this completely breaks down if you include private messages.
This is actually something that's been on my mind recently. If I keep my facebook messages with my girlfriend for posterity (like older people today kept their letters), I wonder if we'll be judged as a shallow couple with nothing to talk about other than daily events. It's just not conducive to much else. In reality, most of the deeper conversations and connections (with friends too) happen in person, with no record to speak of.
In the original Facebook blog post, I believe they reference this when noting that after the switch to "in a relationship," shared posts decline as they give way to in person interaction.
A couple years ago, when I was still on Facebook, I got in this big fight with my girlfriend. I unfriended her on Facebook as a childish act of vengeance, and I noticed that the ads on the side suddenly turned into ads for dating websites. Maybe it was just me, but from the looks of it, Facebook thought I had broken up with my significant other and was targeting ads based on this assumption.
It was then that I realized that every inference they make about you means big business. It means they can now target ads specific to your relationship tendencies on Valentines day, her birthday, his birthday, on your anniversary date, and so on. They can also pick up on events like break-ups and target ads toward you even then. Just another manifestation of Big Data.
What I find most fascinating about these kinds of situations is that it isn't clear that Facebook even "knows" this information, despite taking advantage of the "knowledge": they have a massive statistical inference system that notices certain kinds of events cluster with certain kinds of behaviors, leading to differing reactions to potential advertisements; so, while the website clearly has learned something about you, and if it were sentient might be able to have a really scary conversation with you, past asking the question "show me users who are likely to click on dating ads today" (which is a sufficiently different question as to require its own inferences and leading to its own errors) a developer at Facebook might not have "access" to this knowledge any more than a brain surgeon can use your brain to figure out who your mother is. In fact, these data scientists working their way through the data themselves are only now "discovering" something their ad engine already "understood" a long time ago, and while them knowing it explicitly might help them improve the UI of their website, it likely won't help their ad optimizer as the optimizer might already be "smarter" than they are.
Its starting to sound an awful lot like when skynet wakes up it won't be a terrifying overlord, but rather like a crazy old lady that no one really understands who sells kitsch and loves cats.
Interesting point. it's not necessary that these behaviors are specifically programmed into the ad engine, rather it might have learned them on its own from the enormous amount of data it has.
That's actually a big problem in machine learning. We can make AIs that learn from the data, but trying to reverse engineer the models that they learn is extremely difficult.
I remember hearing about a scientist that used symbolic regression on some of his data sampled from microorganisms of some kind. Eventually it discovered an equation that described the data perfectly. But he can't publish it because no one understands it or can figure out why it works.
Thank you for the very illustrative explanation. It's fascinating to think about how smart the ad optimization will be after another decade, after the size of the training data set gets exponentially larger and the algorithms become more and more refined.
What you're experiencing post-breakup is merely Facebook normal mode, which your girlfriend was protecting you from. (I'm aware of the targeting rule due to work on Internet advertising and aware of Facebook normal mode due to being a single 20-something male for the majority of my Facebook history.)
There's nobody at Facebook who said "Single? Hit him with all the dating ads!", it's an emergent consequence of that policy rule, auction-esque mechanisms for ad placement, and the measurable behavior of most people demographically similar to you, which is that they click on cleavage but buy virtually nothing as a result of Facebook ads.
That's interesting. Still, it doesn't sound too unlikely if Facebook allowed targeting ads to those who just broke up.
For example, when you target your Facebook ad you can pick those who have just traveled somewhere, or even “intenders”—those, I assume, who only think about traveling.
You can also target your ads to people in long-distance relationships already. There seems to be no way to specify that explicitly in your Facebook profile, and yet Facebook provides the category for advertisers.
The first Facebook page I saw immediately after marking myself as "engaged" started showing me wedding related ads, and I fully expect "married" in a month will mean ads for home buying and baby clothes.
It was a nice change from "we want smart, educated Asian women to donate eggs for $10k+" though.
I would have loved to have the ads "we want smart educated black men to donate sperm for XXX", then again, this might be one of those, 'careful what you wish for, you just might get it' sort of things.
I'd rather have ads that might be relevant than ones that are not.
This is preposterous. You would rather be manipulated better? Seriously... This is not a teneble position. Most people would prefer white noise. What you are espousing are simply the talking points of the advertsing indusrty. This is like a congressman thanking the lobbyist for "educating" him on the issues.
You're missing his argument completely, which is that targeted advertising has upsides for both the advertiser and the consumer. While I prefer no ads over ads, given the choice I'd prefer relevant ads vs random ads.
That being said targeted advertising infringes on your privacy. Rail against it for that reason. Don't make false assertions that people would prefer random advertising than advertising that they may actually be able to use.
You are assuming a benevolent advertiser that uses knowledge about you in a way that is aligned with your long-term interests.
If we assume that a white noise advertising is less effective than a targeted advertising then it is in the interest of most people to prefer the former because a benevolent advertiser is a rare bird in real life.
What he is saying is that advertisements manipulate you (likely without your knowledge) and that this is a net cost the the consumer (not to mention even dirtier ads for scams and malware, that's a different subject.) The other kinds of advertisement, the ones that actually inform you of something that is useful to you, do exist. But how much of a benefit are they? I don't think I've ever clicked on an internet ad or bought something because of one.
You are missing a very big point of advertising, which is not just to convince you the product is better than the competition, but that the product exists. In your example, it's like a congressional representative thanking a lobbyist for alerting them to a problem (which presumably they would then educate themselves on). I see no problem with that.
I admit, I would rather the information come in a less obnoxious form in many instances, but let's not kid ourselves that ads have no informational content. Given that they produce some level of useful information, increasing the the level of usefulness without reducing privacy seems a laudable goal.
but let's not kid ourselves that ads have no informational content
Every piece of knowledge has an opportunity cost.[1] To make this point more explicit, the cost is asymetric. Advertising is no different than a drunk on a subway or a schizophrenic homeless person shouting nonsense to a passerby. Yes, they are communicating information. Let me put a bunch of this "information" in front of an advertsing executive's home in Atherton, and watch him call the police and have his lawyer write a letter to the housing association to change the zoning code. And then, the circle of life (and information, and special interests) will be complete/. =D
But the cost is already paid. We aren't talking about increasing advertising, we are talking about increasing the usefulness of advertising. Interspersed with that schizophrenic's nonsense is the useful tidbit reminding you it looks like rain, and there's a cheap umbrella a few steps ahead... Not to mention in-between his ramblings he does a pretty entertaining Hamlet. ;)
Everyon understand's advertising makes other people money. It does so at the expense of polluting the information flow of other people. That is a market externality being exploited by one group at the expense of another. One person's eye0sore is another person's "information". But they have zoning laws that dis-allow LA style billboards in expensive neighborhoods...for good reason. The "information" has a negative externality on real-estate costs (aka: its an eyesore). And just like IRL, online advertising is the same way. Who makes the rules is just a matter of people's self-respect. It turns out that wealth white people find it degrading to be subject to advertising, so they put all of those ads for menthol ciggarettes in and alchohol in the black people's neighborhoods.
I'm honestly unclear why you keep trying to shift the argument to whether advertising is a good or bad thing. That wasn't the assertion (or statement of preference, if you will) of the comment you first replied to, and it isn't my assertion one way or the other either. I believe I've made my position clear on whether making existing advertising more useful for the targeted individuals is a good thing, given that privacy is protected. You've made your opinion on some other issue very clear.
And just like IRL, online advertising is the same way. Who makes the rules is just a matter of people's self-respect. It turns out that wealth white people find it degrading to be subject to advertising, so they put all of those ads for menthol ciggarettes in and alchohol in the black people's neighborhoods.
So what's the online equivalent to wealthy white people keeping large obnoxious advertising out of their neighborhoods, and the advertising dollars being redirected towards less affluent or influential people? Who are the elites in this situation?
User's time and bandwidth are valuable commodities. To safeguard these resources, people engage in social activities which exclude people likely to waste their time. There are entire industries that peddle influence through social engineering to penetrate this approach. Surely I don't need to spell these out ... ;D. But you don't see much of the activity you are interested in here at HN, and when you do the evicence suggests it is the exceptions that prove the rule. And also when those exceptions are not in place it is the typical case demonstrating the case quite nicely.
The issue about "better targeting" is that the delterious elements of advertising do not go away. This is the reason why sometimes we decline to take a phone call in the middle of the meeting. Its not always the targeting, the content, or the message that it counter-productive. Sometimes it is the time and the context.
So it is not just that privacy is protected, or not. Nor is it sufficient that "better targeting" should be an excuse to increase attack surface for placing adds; or that "more secure" should increase the attack surface for data-intensive surveilance and information gathering.
Part of the art of marketing is reaching people in a constructive way. Advertising is not always the way to do that. No matter how "well targeted" it is. That is why we have PR and lobbying, informal networking, social circles, and country clubs, alumni networks...etc. And why we don't accept advertisments in front of phone calls like we do at the movies. Etc.
You're fighting an uphill battle with your argument, because I'm quite sure I've never (nor have many people reading HN) encountered targeted advertising based on data mining on sites such as Facebook, that has resulted in tangibly better ads.
I simply have never had my life enhanced by them in the slightest. Curated ads on blogs with content I appreciate are different. But that's editorial curation.
Curated ads on blogs with content I appreciate are different. But that's editorial curation.
Which is fine, and is in line with my point, which is that given ads, making them more useful without increasing the negatives, is a net win. I'm not arguing that we have seen that, just that if we could, that's a good thing.
Distilled this down to its most basic form, names on the front of shops are advertising. To someone looking for a specific shop, they serve an immediate and useful purpose. To those just trying to enjoy a stroll down the street, they may seem annoying eye-sores. At any future date, those roles could be reversed.
Like most things in life, advertising spans a spectrum, in this case from almost entirely useful with few negative side effects (reasonably represented shop names), to entirely useless (long boring message about a product you couldn't use, event if you wanted to). I see no reason not to try to eliminate as much of the latter as possible.
IMO "informational" ads are few and far between. Most ads are for manipulating people in some way. Not to mention ads for complete garbage like scams and malware.
This is preposterous. You would rather be
manipulated better? Seriously... This is
not a teneble position.
I watch youtube videos about electronics [1]. Some of the content I watch is created by full-time video bloggers who rely on ad revenue to keep producing content. I figure skipping all ads isn't universalisable, so I watch ads from time to time.
Sometimes I get interesting ads which do things like tell me about the new features I'd get if I upgraded my 10-year-old CAD software to the latest version, or if I brought a newer digital storage oscilloscope. These are things I could conceivably want, and that I'm interested to be told about.
Other times I get videos for "latest horroscopes in hindi, the dynamics of capricorn" which there is no chance of me watching ever.
Is it so hard to believe that, given I am going to see adverts, they might as well be adverts with some relevance to my interests?
>I watch youtube videos about electronics [1]. Some of the content I watch is created by full-time video bloggers who rely on ad revenue to keep producing content. I figure skipping all ads isn't universalisable, so I watch ads from time to time.
This is an interesting concept, subjecting yourself to manipulation as a form of payment for a service. Of course if the ads are genuinely useful some of the time, it might be worth it. Or if you believe that ads don't affect you. And that you are somehow tricking companies to waste money on ads that are ineffective.
But there is an argument to be made that advertisements do affect you without your knowledge and that watching them is a net loss.
Absolutely I'd rather be "manipulated" better, if you count advertising that way. I'd much rather be shown ads for things I might actually want than stuff I have no interest in at all, and I say that as someone who has actually purchased things off of Facebook ads. If the ads going to be there regardless, might as well work it for the benefit of both me and the company presenting it.
This assumes you already want to buy the item in question, and advertising just helps you find the best place to buy it, or the best model to get. Consider the other side of the coin. Let's say Facebook shows you products that you don't already want, but based on your age, interests, and past behavior, you have a high likelihood of wanting. The barrage of ads you see convince you to buy that product. In this case, making the advertising more "useful" through better targeting doesn't save you, the consumer, time and money. Just encourages you to buy something you we're perfectly happy without. So is this the kind of advertising you'd like more of? The kind that sells more products for the company, but doesn't help the consumer make the best purchasing decision?
Now I'm not opposed to advertising, and I think advertising can benefit companies and consumers. But we need to start a debate about the usefulness of advertising to separate good advertising from bad. Please pick apart my example above. I'm sure my argument has weaknesses, and I'd love to know what they are.
I"d like to add that I'm also OK with Facebook/Google/whomever making money off that. I think it's just a spectrum of perspective. One side is privacy concerns, the other is the win/win/win where I see ads relevant to me, facebook makes a little money [in exchange for providing a free service that I value], and the advertiser makes a sale for promoting a product that I want.
I don't know the right answer, but I think there's a balance to be struck.
no it can't. Not unless 'privacy intact' means "people with access to the database will never use it for unintended purposes and it is completely secure from being broken into" both of which can't happen.
Wait, if you broke up with someone a few days ago and are now seeing dating ads, how could your privacy have been kept intact? This implies that an ad network has knowledge about your love life.
The assumption is that you've changed your relationship status, and that's how they know you're now single. Telling Facebook that you're single, and them using that information to present an ad for a dating website is not an abuse of private information. You've freely given a first party some information, and they've used that information to display an ad from a 3rd party. There is no abuse of privacy here.
And yet Facebook ads don't seem to be that effective - certainly nowhere near where Adsense ads are, which appear because the user would be looking for dating sites, as in your case.
"Facebook might understand your romantic prospects better than you do."
This article makes me angry.
4 years ago, I met a girl and we added each other on Facebook. For a year, she saw my posts, but Facebook was hiding all her posts from me.
Ultimately we ran into each other again and fell in love. But it really sucks that I didn't see that year of her life while it was happening. Facebook simply decided not to show her in my timeline.
I wish that my friendships and relationships were not judged by an algorithm or censored without my knowledge. Machine learning et al are great for numerical tasks, but let's not try to treat emotions quantitatively. I would hope that we have a better sense of our emotions than computers reading just a fraction of my communications with someone.
I am 30 next month, and I don't consume social media at the same rate as others. I still enjoy meeting up or ringing people! My brother who is 37 dines at the fast-food serving social media restaurant far more than I do. He practically drives through! Wittily (and entirely seriously), he asked me "Do you use Twitter? It's like free text messaging!".....
Perhaps I am a bit of a stick in the mud? Besides, if I liked a girl as the parent said, I would much rather be with her (literally!) than feeling cheated that I couldn't observe portions of her life that she chose to share as served to my handheld device. It's like viewing life through a tiny viewfinder instead of looking around a bit more.
What exactly are those data points? I doubt they are individuals, they wouldn't wrap so tightly around the curve around such a tiny scale (between 1.5 and 1.7 timeline posts per day).
This looks like a very small effect but one with a very distinct form so I'd be curious to see the details.
I was thinking along the same lines. With millions of users, I suppose it makes sense that you can see "average" number of posts per day change when a relationship changes, but at the individual level how could the algorithm tell if it is a change in relationship or somebody was having a good week and felt like being chatty (of course these options aren't mutually exclusive! :) ).
Exactly my thoughts and I don't see a reasonable person/group filter to construct averages without introducing some kind of bias. And with the graph's axis starting at a non-zero value, I'm not convinced that the goal was to provide an objective view on the effects of relationships.
If Facebook can predict when a relationship is due to begin, why not just set the relationship status automatically? They must be doing this internally even if they don't display it.
Or how about they notify all your friends: "John is sharing 1.65 posts/day with Janet but only 1.32 posts/day with his girlfriend Jane. Click here to suggest he dumps Jane and starts a new relationship with Janet."
That's undoubtably their end goal. Even if it's not made public, they would then use it to target ads. Basically, if they can accurately predict human behaviours before the humans can, they can target the ads at the most effective point in time.
I wonder how much fuzzy logic they keep behind the scenes. e.g. John is p=.95 in a relationship with Jane, and p=.1 relationship with Jill, and negligible for all others.
I read somewhere that internally at early Facebook they did calculate rough percentages of likely relationships, but that they stopped because it could be publicly damaging. I'm pretty sure this was the same article that talked about how they also assigned a 'human' percentage to each account based on metrics to try and weed out spam.
What exactly kind of graph is this? I see three plots: a blue line, a thick purple segment and black speckles. None of them are labelled.
Is one of them a mean, for example? What is the variance in these data? My gf and I only post on each other's pages once per month, max, for example, yet we ust be part of this "study." The variance is especially important because the six-month span from top to bottom (1.66 - 1.54 = 0.12) is large compared to mean of the whole graph (~1.60).
What kind of selection criteria were used to create this graph? How many couples? What ages? What countries? (I don't accept "worldwide" b/c I don't accept that people of all countries post at the same rate).
This "information" has all the marks of Facebook trumpeting their macho data muscles to impress advertisers and the Atlantic flexing their paranoia muscle to stoke reader's outrage. I suspect those black dots are simply graphical saccharine to give the aesthetic appearance of "data."
Exactly. It's just a minor effect that Facebook can detect using their massive database and use to convince advertisers to pay a premium for, even though there is no evidence of how good a predictor it is. How many people that have a slight increase in their cross-posting actually get into a relationship? Using bad data science used to sell ads.
Mine is just for fun, or rather just to speculate on something interesting from a geeky perspective to see what that perspective reveals about ourselves. Soon after making it I saw a plot some guy made of the number of texts between him and his girlfriend that vaguely resembled my plots. The Facebook plot doesn't continue to a relationship's decline, so it can't compare completely, but it does suggest we can quantify some interesting patterns and see what they reveal about us.
If there's anything most of us would like, it's to understand our passions and attractions to improve these emotions and the relationships they lead to.
It would be sad if the only entities using the data were companies that sell ads like Facebook and Google, and not ourselves.
Also, if you like data analysis like this on relationships, be sure to check out Okcupid's Oktrends -- http://blog.okcupid.com.
Where exactly do you derive your passion / attraction model from? Where's the data supporting this? I must confess, I haven't finished your article yet. But I don't see any hard data besides the aforementioned text messages... a single data point.
I would guess you are correct: that the signal is pretty weak per individual couple (to the point of not being usable) but shows up with this nice tight graph when you aggregate data.
Given that the number of posts dips over time, that's a difference of approx 8-10 posts in that 85 day period after the relationship begins. Seems pretty negligible.
This is selective after the fact analysis. The data of how people behave before and after a relationship positive apex (two people becoming a couple) is irrelevant unless compared with those with a negative apex as well (those that did not become a couple).
In my experience the data presented here shows the need to cut further or obtain additional data. There's some correlation, but the data appears to be a rough aggregate and the absolute magnitude of the movement is small. Also unclear if cyclical or macro effects were removed. Pretty much all the other comments here are anecdotal or attempted to draw conclusions from insufficient data.
This is why pop psychology is generally reviled by psychologists.
I'd like to see a graph of my own personal interactions with each of my FB friends over time. Obviously, I already know what this graph would look like, but it'd be pretty neat (and maybe a little sad) to see my failed relationships juxtaposed with the ones that have lasted.
I know FB is built to create engagement on their site, but I don't have an FB account - the data science team posts are amazing and it's the first content that makes me want to "follow" along. Is there a service to convert FB pages into RSS feeds?
I remember reading on HN that some US supermarket [0] could predict pregnancy based on shopping. So they would start sending baby-care products offers before the couples found about it. They ceased doing that because people found it just too creepy. Any more remembers more?
[0] a really large one, so that I recognised it's brand while never been outside Europe
The store was Target. The story was that a teenage girl was buying things there that was indicative of pregnancy and her father found out through the ads they sent her.
It's simpler than that. It isn't necessarily that the company predicts you are pregnant, they just see that "people that buy this item are more likely to buy these items" and so advertise for them. I believe Target did target pregnancy specifically in their algorithms though.
It would be more interesting to also see the stats about IM's between two people. Also does their study account for all the teenage girls that post that they are in a relationship with their best (girl) friend?
Wow, 100 days, that's over 3 months, what can they possibly be sharing/talking over 100 days? So we are saying this people are "trading" the outcome of 2-3 dates with 100 days of cat pictures sharing? I really don't get the point of this.
an increasing number of posts leading to a relationship and then a drop in messages as the couple is spending time together - OMG GENIUS I WOULD HAVE NEVER GUESSED!!!!
I think a lot of people would have no problem predicting this pattern without looking at any Facebook data. Likewise, the participants and their more observant friends surely know that a relationship is likely to happen well before it becomes official. Relationships mostly follow the same course. I guess it's mildly interesting to actually see it in a more concrete form.
The fact that people make fewer posts on Facebook as they become closer might mean something, in general, about the kinds of relationships that Facebook sustains.
Facebook is the modern equivalent of a Christmas card...It's a way of maintaining surface-level contact with many people: letting them know that you care...but you don't necessarily care enough to catch up through a more direct and personal medium.
Communicating with your closest friends or your partner mainly through a Christmas card would be silly, in my opinion; we tend to choose less distant methods of communication. So, it's not surprising that people would exchange fewer posts with close friends, as well as romantic partners, as they get to know them.
Those are just my two cents, speaking from experience and observation. I'm sure HN has a range of opinions about this, and I'm interested in hearing other people's perspectives.