2
$\begingroup$

Problem:

I want to maximize performance for social media posts by optimizing the time when they are published.

Current model:

X: publishing_datetime, post_attribute_1, ..., post_attribute_n
y: performance

Desired model:

X: post_attribute_1, ..., post_attribute_n
y: publishing_datetime

The desired model should predict the optimal publishing_datetime for maximizing performance. Once the data can be modeled like this, the problem is solved with a regression neural network.

What I've tried:

Filtering the posts with above-average performance and using their attributes and publishing_datetime to form my desired model.

This is not ideal as a lot of data is unused and posts with particularly great performance influence just as much as posts with barely above-average performance.

Any suggestion on how to achieve this model transition?

All ideas and alternative approaches are very welcome. Thanks in advance!

Improve this question
$\endgroup$
2
  • 1
    $\begingroup$ Is using the existing model and „fuzzing“ possible publish times to optimise the predicted score a way to go? $\endgroup$
    – AlexR
    Mar 11, 2018 at 21:44
  • $\begingroup$ Yes, that would work! I'll leave it as a last resort though, as fuzzing through timeframes would require extra processing. I'm playing with collaborative filtering, where post_categories are "recommended" to each timeframe. $\endgroup$
    – Jorge Barrios
    Mar 12, 2018 at 3:34

2 Answers 2

Reset to default
0
$\begingroup$

You might be confusing target with feature interpretability.

You probably want the target to be performance and datetime as a feature that could be interpretable.

If datetime is encoded in epoch time, it might be to sparse to be useful. You might want to construct many datetime features, examples include: time of day, day of week, day of month, and day of year. The result would be a model that could predict how different times effect performance.

Improve this answer
$\endgroup$
0
$\begingroup$

One way is to construct the problem as a classification problem, where you break the week into 30m slots.

You'll need to carefully define the label of success - after normalizing the channel activity and looking for a high z-score.

This way you can still use your desired features.

To further improve the model you'd like to use seasonality data and holidays. (you can get those features from prophet

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.