JMP for optimizing UXR website (split-plot DOE ?)

Emmanuel_Romeu

Hello,

I have a lead (re-seller of chemicals in aerospace) who is wondering if their company can use JMP to optimize their website.

I remember a similar internal presentation (as well as a customer presentation at Discovery Summit) about building a DOE (split-plot design ?) and analyzing clicks or time.

Does anyone can direct me to anything I can share on this topic? 

Thanks for the help.

russ_wolfinger

Hi @Emmanuel_Romeu ,   

One idea for DOE is to design web promotional campaigns kind of like we often do with fun experiments in the JMP division.  

Wanted to also mention this has been a hot topic in data science ever since the Netflix prize and data from 2006.   Good buzzwords to search are "click-through rate" and "recommender systems".  There are a bunch of datasets and Python notebooks online, and you can even run the latter for free with GPUs on Kaggle.   These are very well studied machine learning problems and used heavily in websites for entertainment and shopping.   

The original problem is formulated as a "user-item matrix" with users as rows, items as columns, and data being some measure of activity like a rating or click-through-rate.    The matrix tends to be large and sparse, and initial breakthroughs were made with fast, low-rank approximations.  These evolved into machine learning embeddings and methods like factorization machines.  Nowadays companies like Amazon are performing a massive number of blazing fast recommendations behind the scenes in real time. 

Crazily enough, I was just recently trying the MovieLens 100K data when doing initial testing for the new tdl.na.sas.com Windows server.   If you remote desktop into tdl, there is a table C:\Public\movielens100k_rating.jmp with an embedded script.  The key for that one with the Torch add-in is to jack up the Batch Size to something like 32768.