Distributing online traffic to the optimal paths - something for deep learning?


I’m in the process of studying machine learning at Edx and fast.ai. I’m currently getting a grasp of things, but look for problems that I’m familiar with to better use it in practice.

My wife works in online marketing, and an interesting system used by many companies is how to distribute the traffic and/or change bids for a click based on the data.

So I imagine it like this, a click will have independent variables such as:

ID, timestamp, os, browser, isp, banner ID, IP/CIDR, websiteID, tag, referrer, campaignID … ParameterN

Then based on this, you would route traffic to the best converting “path” for this segment, i.e.

Timeofday -> OS=Android -> campaignId=abcd1234 -> Browser=Android Webview

An example: https://doc.voluum.com/en/auto-optimization.html

On a high level, how would this be attacked?

I learn a lot better when I can connect knowledge to real problems. I guess what I’m struggling with is to take a problem like this, and attack it when only having done online tutorials. Is it something that could be done in deep learning, or more of a statistics problem? Any pointers. I’ve been looking for parallell examples i.e. on traffic lights, sales optimization in stores, but none of it seems applicable to this problem.

Thanks in advance

I guess what I’m looking for is recommender systems (recommend a path/url) with multi input (several variables). I found some papers, will ask a friend in uni to grab them for me. :slight_smile:

Is this on the right path? It’s the closest examples I’ve found in similar industries.

Would love to discuss this with you gmovr