Twitter's graph (2012) | Hacker News Reader

Twitter's graph (2012)

dcurt.is

46 pointsjimsojim22 comments

> Twitter has an enormous advantage over Facebook in one key area: while people on Facebook tend to friend their friends, people on Twitter tend to follow their interests. The social graph that makes up Twitter is worth far more on a per-account basis because it is directly monetizable in a way that Facebook’s generally isn’t – you can show prophylactic advertisements to Twitter users based solely on the people they follow, and probably get a much higher rate of interest. Compared to other social display ads, Twitter ads, it is rumored, work extremely well.

So, clearly this is from 2012 and the thesis is 100% incorrect (in fact, pretty much inverted). The interesting thing is that this is pretty much how the general populous still thinks Facebook does it's ad targeting. The author massively overlooked machine learning. I don't have first hand evidence, but I can only guess that Facebook is crawling every piece of data possible (pictures, messages, etc) and then draws conclusions of my interests there. In addition to this, all FB user info is being hashed and compared against consumer spending data (e.g. DLX and Acxiom).

The level at which FB and Google can target us is staggering. There is a reason these companies make so much money.

> The level at which FB and Google can target us is staggering.

I'm so glad I deleted my facebook account years ago for this fact alone. I don't need to give all my information for free to facebook to use what is essentially IM.

Twitter still has higher ad conversion rates.

Machine learning can only do so much. Human editing will always win over machine algorithms, and Twitter users curate their follows. Also, machine learning only improves conversion rates by 2x. Meanwhile, I've seen a 10x conversion rate improvement when hand editing ads on a site. Machine learning ads are why you have out-of-context ads that destroy a sites user experience.

> Twitter still has higher ad conversion rates.

Source? Genuinely curious, about to kick off a campaign on Twitter in an esoteric area (triathlon) and can't imagine it will be a higher ROI than FB.

> Also, machine learning only improves conversion rates by 2x. Meanwhile, I've seen a 10x conversion rate improvement when hand editing ads on a site.

I'm not sure you're following my line of reasoning here (or maybe I'm not following yours). The machine learning becomes useful when I try to target people's behaviors. For example, I say "Facebook I want to run an ad against all people who actively run and get them to buy running shoes". Facebook is going to optimize to serve up ads to people who (1) run (determined by messages they send to people "hey great run, I did 5km last week" or integrations with RunKeeper, Strava, etc, or maybe I have pictures of me running a race) (2) click ads (easy enough) and (3) will purchase online (determined by conversion pixels of previous campaigns). This is a gross oversimplification of course.

If we're on the subject of citing sources, what are your sources for saying that:

> the thesis is 100% incorrect

and guessing that:

> Facebook is crawling every piece of data possible (pictures, messages, etc) and then draws conclusions of my interests there. In addition to this, all FB user info is being hashed and compared against consumer spending data

That's a pretty sweeping statement. You'd be surprised at how naive (and distinctly not-amazing) the machine learning algorithms used in-production at places like Facebook can be (in-production, as opposed to cool research papers that aren't actually used in-production).

Going off of your simplified example, there are tons of contexts in which someone might say "run" without meaning anything related to the athletic activity. That's the definition of shitty targeting (false positive).

For all their ability to target us, the ads I get on FB & Google are woefully untargeted. I wish I only saw ads I was interested in, but the hand curated ads from The Deck or on podcasts I listen to are far more relevant to me than Google or Facebook ads are. (Mail Kimp FTW)

If that means Facebook & Google have more room for future optimization, awesome... but I fear they really make money off sheer volume of ad inventory. Watch the Facebook ads and how they refresh every couple of minutes - that isn't about extra targeting or clever machine learning, that's just extra inventory of ads charged per impression.

I find it odd that somebody would spend half an article talking about how users consume Twitter as entertainment then claim that it's the developers who will determine the platform's fate. It seems to me that if the content is what keeps the average user around, it's most important to keep the content creators happy. Twitter hasn't had a stellar record with this so far (especially w/r/t harassment), but they have succeeded in not alienating lots of giant celebrities, which seems to me to be the only thing that could conceivably kill today's Twitter within the next 5-10 years.

> where the developers go, the users will follow.

This doesn't seem to be true. It's true of real platform plays, like a desktop or mobile OS.

Yup, and not even with desktop / mobile OSes. Generally it's the other way around. Where the users go, developers will follow.

There are cases where the relationship is important, like in the early stages of a product - where in some scenarios you need both. And yes, if you need developers, and developing for platform x is hard, and y is easy - all else being equal y would win.

>When the disruptive competitor comes along – when, not if – who are the developers going to side with?

There can't be more than 50,000 devs who used the Twitter API. I definitely don't think they're going to be the deciding factor.

That being said,I would never use the Twitter API based on the horror stories I've heard.

2012 was before Twitter went thermonuclear with their API limitations. (back then, you could even access the Twitter API without authentication!)

> [...] what value does Twitter have, other than a more restricted set of content? What about App.net? Twitter is in an even worse position than MySpace to fight off a disruptive competitor.

Did this article imply that App.net was about to kill Twitter since App.net was open?

Heh.

I think the point is: If Twitter had remained open, it would have been easy for users to export their graph and take it to competitors with similar features. Twitter closed their friend finder APIs to prevent this.

I don't know if App.net could have ever killed Twitter, but Twitter closing made it harder for users to transition to App.net in the first place.

Not sure if I understand correctly, but doesn't Facebook have the "interest graph" completely built in through facebook pages? In other words, just as Twitter knows a user's interest through the hundreds of people they followed, Facebook knows a user's interest through the hundreds of Facebook pages they liked/followed.

Isn't that completely the same or isn't Facebook even better in that respect, because Facebook pages are much more often real interests, whereas Twitter users that are followed are more often people instead of interests?

When this was written in 2012 Facebook didn't have the concept of following at all, only friending. It's easy to forget that they subsumed most of the good things about Twitter.

It's easy to forget, because it never happened. Following is not "a thing" that's popular to do on Facebook. Likes and friending are really the dominant interactions.

I'll have to disagree - we do Facebook page monitoring and I often see random brands with millions of fans per country. I often thought the same thing because I never friend anything, but I am clearly not the majority.

That doesn't make it the nexus of activity on Facebook.

If friending disappeared, Facebook would cease to exist as a product. If liking disappeared, a huge chunk of activity (40+%?) would disappear. If following disappeared (you could still friend or like), I don't think many people would notice.

> That doesn't make it the nexus of activity on Facebook.

That wasn't my contention - your contention was:

> Following is not "a thing" that's popular to do on Facebook.

And yet, the hundreds of millions of people who have followed brands on Facebook would indicate that it is popular. It can be popular without being "the nexus".

I said it wasn't "the thing." I meant it wasn't very important. You're also talking about likes. People "like" brands. They don't follow them. Following refers to people following other people, without friending them - this is Twitter.

I used to follow various things on Facebook. It's pretty normal and a good piece of functionality for many people. However, I did recently unfollow all friends and groups leaving me with fewer ways to get distracted by FB.

> I suspect the reason that Twitter is cutting off apps from using its “friend finder” feature is because...

How did this happen? I've been using Twitter API these past years but I'm not aware of this "friend finder" feature. Getting friends/followers is still there in the REST API.

Reply on news.ycombinator.com