What structure should have a prediction neural network?

Hello!

As the thesis decided to create a neural network, which (theoretically) could predict the dollar on the basis of news articles. I understand that it is impossible in principle, and the result will be close to randomname, but it's a degree, so the result is not so important :) I Want to ask for advice. I must say that about neural networks, I started to read just a couple of days ago, so please do not throw sneakers.

Now I've spars articles about 147000 for the last 9 years, highlighted key words, lexically has processed, cleaned in my opinion unnecessary. So from 90k key words I got about 40K.

As for neural networks, I knew only approximate the principle of their operation: the input we submit some data, the output is the probability of a positive or negative response. At least, this principle has been described in the guides that I read) But there was considered only networks with 5-10 inputs, each input has moved 0/1, the output is the probability unit. I figured out how to create such a network and how to teach them. But, unfortunately, articles that describe neural networks that could solve my problem (or close to it), I have not found.

At the moment I'm having 2 questions:
1. How to create a neural network that could process keywords? The only thing running through my head was to create a 40K input ports to the network. But it is logical that it is wrong. Moreover, at a certain point, keywords can be added.
2. How to make a neural network were given not the probability of occurrence of the event, and the row direction (the rate goes up, strongly up will remain the same down, etc.)?

Not asking you to fully describe the algorithm, could you tell me please the names of the algorithms, techniques, etc., articles that can help me :)
Thank you in advance
March 19th 20 at 09:20
2 answers
March 19th 20 at 09:22
As the thesis decided to create a neural network, which (theoretically) could predict the dollar based on news articles
it doesn't work

Not only to this currency, but in General

Also don't think this idea you came first to mind

Also had the feeling that you don't understand how ML works in principle in the real world

But, unfortunately, articles that describe neural networks that could solve my problem (or close to it), I have not found.
Those you have not found how is fact extraction, sentiment analysis, samples to write a program for prediction of courses and find the formula which is the naive forecast buy/sell on investing.com

My advice is to change the subject while there is time

How to predict the ruble
If I was 10 years ago, when asked a similar questions on programming and had listened to such advice
My advice is to change the subject while there is time

I wouldn't have become a programmer :) If you in fact have nothing to offer-why flood? I understand that you are very cool date sientist, and know much more than I (especially since I know 0). But in fact you said nothing and, apparently, almost did not read my question.

1. I did not ask to write the entire algorithm for me, asked only to give the keywords
2. I know that to predict it will not work (which I also wrote) on a huge variety of reasons, but it's a diploma and even a negative result.

Also don't think this idea you came first to mind

Oh Lord, I'm at least somewhere in your question claimed to be unique?) If the solution to my problem, I would happily have stolen for diploma-less work would be.

naive forecast buy/sell

If I have correctly deciphered what you were trying to encrypt to seem smarter, I something similar was done in the past :) But I have set myself the task of prediction, not starting from past values, and other factors, even if it is to no avail.

And, by the way, in my experience, people who yuzayut English words where it is logical to use a Russian -- usually are verbiage that attempt to seem smarter. In any case does not apply to you ;) Though in fact, as I said above, you said nothing....... - Rey.Ortiz96 commented on March 19th 20 at 09:25
@Rey.Ortiz96,
You didn't read my answer
Try again

But in fact you said nothing and, apparently, almost did not read my question.
fact extraction
sentiment analysis
sample to write a program to predict courses
formula which is the naive forecast buy/sell on investing.com

Oh Lord, I'm at least somewhere in your question claimed to be unique?)
I hinted that if a solution does not exist, does not exist and its realization

If I correctly decoded
no, I pointed out key words
You have to paste them in Google

And, by the way, in my experience, people who yuzayut English words where it is logical to use a Russian -- are typically the verbiage

experience
are the ranters
- Madie.Sauer commented on March 19th 20 at 09:28
March 19th 20 at 09:24
In Finance there's this thing called "technical analysis". There predict the course without any outside sources, purely on the history of fluctuations.

The words\algorithms for working with text: word embeddings, RNN, LSTM, CNN, Transformer (topic now)

But I'd still tried more close to the ground task. Last year there was quite a curious architecture\art Neural Ordinal Differential Equations, which including you can try to apply to exchange rates\shares.
word embeddings, RNN, LSTM, CNN, Transformer
and there are materials on the given basvurdum with examples of the profit from their fingertips?
Or the names of the researchers, about who is able to write? - Madie.Sauer commented on March 19th 20 at 09:27
@Madie.Sauer, materials on RNN, CNN? - Edgar27 commented on March 19th 20 at 09:30
@Edgar27, Yes - Madie.Sauer commented on March 19th 20 at 09:33
@Madie.Sauer, well, about now, in any course on deep learning tell.
The same transformer is https://arxiv.org/abs/1706.03762 https://arxiv.org/abs/1801.10198 https://arxiv.org/abs/1810.04805 - Edgar27 commented on March 19th 20 at 09:36

Find more questions by tags ForecastingNeural networks