/biz/ - Business & Finance

>>26255075
>Update: The Wojak Index has been discontinued. Because I don’t wanna pay $5 a month for a server and $5 a month for the screenshooter service

fucking jew

>>26255075
>comment on the bottom
Aniki...

I can make this if you fund me
or help me decide how to monetize this?

>>26255290
it's not always about the money. it's about sending a message. i'll do it.

>>26255387
what message?
inb4 jeets just manipulate it

>>26255290
i'm not funding anything you can literally throw up om your VPS anon

>>26255387
based anon
let me know if you need a domain
i have one reg'd for a year for a project that didn't pan out

>>26255469
>i'm not funding anything you can literally throw up om your VPS anon
what is r&d

>>26255387
>>26255290
>>26255469

How about measuring the ratio of pink/green and chart it? Strong pink = sell signal, strong green = buy?

>>26255597
Also needs to measure black/brown for bobo.

Perhaps a better approach is to train neural networks to identify wojak/pepe/bobo? I'm sure some autists have sorted reaction image folders already that can be used to training.

>>26255526
OK, i'll bite
$100 in ETH to whoever shits out this CS 101 tier project

>>26255651
post your wojacks/bobos/etc. here for training data /biz/fags

>>26255075
sounds like ml image clasification would be a really good choice for this and I was looking for something to deepen my understanding maybe I will do something like this these days

>>26255703
memeatlas should be fine

>>26255705
>sounds like ml image clasification
nigga just take prominent color as indicator how retardedly fucked is your brain with this buzzword technology

>>26255703
Cases like this one might pose a challenge for a neural net. What message does it convey? How to train a network that can correctly classify this example?

Perhaps remove all examples where more than one character is detected?

>>26255953
More difficult cases

>>26256013

>>26255597
how i would do it

1. scrape images from /biz/
2. classify images into wojack/bobo/pepe/nothingburger
3. track count and frequency history for each image category (hour, day, week, month, etc.)
4. score sentiment using the counts + LMSR or some other scoring rule for each historical period

>>26256044
Maybe we could classify them by hand?

A sentiment analysis on the first post could perhaps increase the accuracy?

>>26255387
Based

>>26256081
It's simple and possibly effective, but there will be significant noise/errors with this approach. And it's all in vain if the sentiment on /biz/ lags behind changes in price, no?

Look at this one, it's as if someone tried to make one that would fool a neural net.

>>26255597
>Strong pink = sell signal, strong green = buy?

This is how I know you're a newfags and you'll never make it. It's literally the other way around.

Consider this free life advice:
You buy when biz goes AAAAAAAA
You sell when biz goed OOOOOOO

>>26256214

>>26256096
it should be pretty quick to train an accurate classifier for characters using OpenCV or some other toolkit

we should indeed consider post sentiment (all caps, slurs, etcs) since the character in each post does not necessarily correlate witb the sentiment

>>26256214
that is a fucked case

but we don't need to be 100% accurate
+ we can weigh counts by confidence

>>26256235
I agree, but I would add that the AAAA and OOOO must have been consistent for a while so it's a trend. /biz/ is flaky, sentiment turns on a dime even with small price fluctuations.

>>26256287
A naive classifier will surely not be accurate though, given the examples I've posted so far?

>>26255387
Sometimes it's not even about sending a message. I'll pay everyone in this thread to make this project. It's a great opportunity.

>>26256164
maybe we could do said approach all at once for each period to create multiple indexes? each increasingly more volatile + tracking a smaller period of time (down to like 10 minutes or something)

>>26255190
lol people claim to be millionaires on this board but dont pay 10 dollars a month

>>26255075
I spend all day here anyway, don't need an app or whatever to tell me if it's pink or green in the catalog.

>>26256433
ok bos on it will create bread when done

>>26256447
Yeah. On the other hand if we sample frequently and track continuously and serve the data through an API, anyone can analyze it and use it however they want.

>>26255953
>>26256013
Doesn't need ML, just a threshold for proportion of pixels in the 'pink range' for each image to call 'pink images', then proportion of 'pink images'/total images. Bobos don't count and shouldn't be counted.

>>26256416
i'm thinking it might be accurate enough? (since we only need to track general sentiment + not everyone posts sped versions) but it would definitely need some testing and experimentation

>>26255597
ngmi

>>26256572
Pic related, positive sentiment that would be classified as "pink".

>>26256526
it looks like if we're doing the windowed approach we would have the data anyways to expose this api
so why not do both :)

>>26255075

I already have one made. As others have mentioned before me, this is a fairly CS101 type project.

Not sure why I would share it with lazy fucks on here though

What you need is the hash signature of each pink wojack images. Then you can compare the prc. Easy

>>26256881
^ this is a valid approach as well

>>26256463
You can get 0.5 of LINK for this kind of money.

>>26256615
You're probably right. I'm pretty sure that 90%+ of the reaction images that are posted are reposts, so if they were classified once the the maintenance work needed to classify original variations would probably be minimal. Then again for a project like this it's probably not even worth the effort.

>>26256722
Yes. However, if the index runs continually then the false-positives provide baseline data. It's the change compared to this baseline in a pink-fields scenario that's significant. Calling of each image doesn't have to be 100% accurate.

>>26256881
https://en.wikipedia.org/wiki/PhotoDNA

Maybe something like this can be added to increase the robustness against small alterations.

make a discord for the projects ill fund too

>>26256935
i'm sure there is some cs tard lurking here on /biz/ that needs the (((experience))) to look for a wageslave job

if you put some modicium of sophistication into this project you can put this on your buzzword resume as 'Data Engineering' or some shit like that

to all college students: i know you're not spending that much effort at Zoom University, so why not work on this meme project and further some skills? :)

>>26256869

Also, it was mostly because I was curious about which wojaks were out there. If you think this will make you money you're fairly retarded and I'll save you some time: set a price alert with the coingecko app for when BTC moves 10% in a certain amount of time, and voila, same effect, 0 effort.

>>26257172
yes u really hit this is me :)

>>26257014
Fair enough, my gut feeling is that it would probably be accurate enough.

>>26256869
>I already have one made.

Can you share your data set? (:

>>26256463
>I didn't get rich by writing a lot of checks

>>26257216
yeah, this will 100% not make you money
if you are doing this it will be purely for the lulz

>>26256081
They all seem to use the same shade of pink, maybe you could do something with that to avoid image classification. Memes evolve enough that many would slip through the cracks of your scraper