Someone on reddit used machine learning to estimate subscriber number

[yqvfmj]

I estimated subscriber numbers using Google trend data and machine learning, here are the results.

From the OP on Reddit, if anyone is curious what he did.

[Deleted]

To the extent people are willing to go just to know how many subs WoW has.
Its kinda funny.
I must know whats inside the box syndrome

[kichigaijin]

To the extend people are going just to know how many subs WoW has.
Its kinda funny.
I must know whats inside the box syndrome

Blizz brought it on themselves by touting their sub numbers, then doing away with it.

[Deleted]

I don't believe that Legion hit 10 million

Why?

Cata did
MoP did
WOD did

Legion didnt?

[Eleccybubb]

People care waaaaay too much about sub numbers these days....

[yqvfmj]

People care waaaaay too much about sub numbers these days....

Of course, they should care because fewer subs mean less income for Blizzard which means Blizzard will either increase the subscription fee or add more microtransactions or cut/remove more contents. If the game dies = all micro transaction items they bought don't exist on the world any more.

[RuneDK]

People care waaaaay too much about sub numbers these days....

it gives an estimation of the state of the game. I think people put TOO much stock into it myself as well...but it's okay to care about the numbers. Especially when a company like Activision is involved.

[RuneDK]

Of course, they should care because fewer subs mean less income for Blizzard which means Blizzard will either increase the subscription fee or add more microtransactions or cut/remove more contents.

they'll add store content...which isn't bad...it's up to the buyer if they find it worth it. Just means even less cool things to get in game by playing...but im okay with it compared to the alternatives....they also canned content when things were going well...Abyssal Maw anyone....remember that was cut for a daily hub...and that was at a time when people were actually enjoying Cataclysm.

[Deleted]

Way too high current subs for BFA. Maybe if the estimation is after 8.1, then probably 2m is true.

[Eggroll]

People care waaaaay too much about sub numbers these days....

They want to know if they're investing their time in a sinking ship or not. Seems reasonable. Would you keep spending your time in a game you knew will shut down in two weeks (exaggeration obviously)? It's also tied to the question how much effort Blizz will put into the game. If sub numbers are low they suspect that less effort will go into the game. In extenion: less game/quality for your dime.

[Selah]

Can someone explain to me why this would be more reliable than some random forum poster putting out the exact same thing? Because, as far as I'm aware it's not. It doesn't have any extra data that the rest of us don't already have.

[Cracked]

Can someone explain to me why this would be more reliable than some random forum poster putting out the exact same thing? Because, as far as I'm aware it's not. It doesn't have any extra data that the rest of us don't already have.

It's not. It's only extrapolation from data, which machine learning methods are extraordinarily bad at.

[Deleted]

It wouldn't surprise me if these numbers are not far from reality.

[cparle87]

Of course, they should care because fewer subs mean less income for Blizzard which means Blizzard will either increase the subscription fee or add more microtransactions or cut/remove more contents. If the game dies = all micro transaction items they bought don't exist on the world any more.

Says they'll add more microtransactions... the week they're removing microtransactions from the store.

Also, wtf is machine learning?

[XDurionX]

Google trends is obviously a very bad way to estimate subs, since they're not related in any way, shape or form. Could as well use "machine learning" (which i assume is nothing but regression) on coffee grounds.

[Deleted]

Can someone explain to me why this would be more reliable than some random forum poster putting out the exact same thing? Because, as far as I'm aware it's not. It doesn't have any extra data that the rest of us don't already have.

I didnt check the entire thread but all i got was this reply from the OP when asked to share his code:

"The code is a bit hacky, but I'll gladly share the data to get you started. Here's a link to the monthly time series. I got the data from MMO-Champion. Save it to a .csv file and upload it to Google Correlate to find the predictive keywords. You will find that on a large scale (2004-2019), wow interest correlates with random things like Facebook, and no so much with wow related stuff. My hypothesis is that over 15 years, the way people use google changes. For instance, Wowhead, Twitch, and YouTube didn't exist at launch in 2004, so queries like "wow quest" or "wow video" must've been more popular on Google at the time. So in order to find the correct keywords, you will have to zoom in and find correlated keywords by time period. Because we are interested mostly is the last bit (after 2015), you can focus more closely on this time period. Use Google Trend to compare keywords and download your data set.



The Idea and the methodology came from the book Everybody lies by Seth Stephens, which I strongly recommend reading. It's a non technical book about the power of using internet searches as data compared to classic surveys."

https://www.reddit.com/r/wow/comment..._trend/edb30z0

[cparle87]

They want to know if they're investing their time in a sinking ship or not. Seems reasonable. Would you keep spending your time in a game you knew will shut down in two weeks (exaggeration obviously)? It's also tied to the question how much effort Blizz will put into the game. If sub numbers are low they suspect that less effort will go into the game. In extenion: less game/quality for your dime.

EQ is on its 25th or so expansion with fewer than half a million subs and has been in that area for over a decade. People thinking the game is going to shut down and all their effort is going to puff into the wind in the time it would take to sire a child and send them off to college is ludicrous.

[Loosecannon]

I Like the idea in practice, but as a hobbyist myself had some questions-

How much Google trend data did you use? What specific data? It looks a little bit like it's slightly overfit to the historical numbers (Just by my eye, nothing else to go on). Also, What algo did you use? Was it an RNN or something simpler?

I could just about buy the legion numbers, but I think the BFA numbers are a little too noisy - Even if they are perhaps in the right area. I'd love it if blizzard gave us the actual so we could see how accurate the model is

[CerealLord]

From the OP on Reddit, if anyone is curious what he did.

I took all the available data points from the quarterly reports and did a correlation search. A few keywords came up highly correlated (~.96), such as "play wow", "shadow priest", "wow guide", etc. It's very interesting to see that even the smallest local peaks (e.g. patch releases) are highly correlated across those keywords.

I then trained a regression SVM using all the keyword trends. The reported error is over a 5-fold cross validation.

The code is a bit hacky, but I'll gladly share the data to get you started. Here's a link to the monthly time series. I got the data from MMO-Champion. Save it to a .csv file and upload it to Google Correlate to find the predictive keywords. You will find that on a large scale (2004-2019), wow interest correlates with random things like Facebook, and no so much with wow related stuff. My hypothesis is that over 15 years, the way people use google changes. For instance, Wowhead, Twitch, and YouTube didn't exist at launch in 2004, so queries like "wow quest" or "wow video" must've been more popular on Google at the time. So in order to find the correct keywords, you will have to zoom in and find correlated keywords by time period. Because we are interested mostly is the last bit (after 2015), you can focus more closely on this time period. Use Google Trend to compare keywords and download your data set.

[Twdft]

They want to know if they're investing their time in a sinking ship or not. Seems reasonable. Would you keep spending your time in a game you knew will shut down in two weeks (exaggeration obviously)? It's also tied to the question how much effort Blizz will put into the game. If sub numbers are low they suspect that less effort will go into the game. In extenion: less game/quality for your dime.

Only play if you have fun and it doesn't matter at all how many subscribers there are.