0
$\begingroup$

I'm working with social media documents.

My aim is to predict number of posts, which contain specified words (terms). I have plenty of data, millions of posts, so, it seems, that model can be efficiently trained.

Again about the aim - I have social media posts for last year (till now) and need to predict the number of posts, containing for instance word "apple", for tomorrow. I understand that the actual number highly depends, but, on the other hand there are some dependencies in dynamics of posts propagation, maybe, something like in stock markets prediction task.

But there are so many methods, I do not know the one to choose for specific task, because each method is good in its own scopes.

Is ARIMAX good for this purpose? Or should I choose neural networks approach? Or something else (ES? MMSP)?

I'm not a scientists, so it's really hard to choose the right way to start.

$\endgroup$
0
$\begingroup$

ARIMAX is the correct approach ... just make sure that the Gaussian Assumptions regarding the error structure are tested and confirmed. Specifically no outliers/level shifts/season; pulses and local time trends remain in the errors for a useful model. Furthermore there is no indication that the error variance changes stochastically/deterministically or in proportion to the expected value from the model and finally that the model captures all the information in your user-suggested X's.

| cite | improve this answer | |
$\endgroup$

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Not the answer you're looking for? Browse other questions tagged or ask your own question.