neural network - Algorithm for cross selling -like amazon people who bought this -

neural network - Algorithm for cross selling -like amazon people who bought this -

new , long time since i've done programming or forums .... however, getting under skin.

i've been looking around @ algorithms used amazon etc on recommendations create around products have affinity ones people have selected - works well.

here wondering....

a - why limited affinity? there never situation product exclusive of original selection , perhaps parallel not product might create sense?

b - why neural network not create sense? not work provide link or end few product have low weighting , hence perpetuates non selection?

thanks views.

james

question a: not need limit affinity. however, need "package up" other pertinent info in way can nowadays algorithm. should read on "association rules", "frequent itemsets" & recommender algorithms. of these algorithms analyze transaction info , larn rules {peanuts, hot dogs} = {beer}. going away affinity can produce multiple sets cut down beer {alcoholic beverage} utilize multiple frequent item sets, @ different levels of specificity, , sort of ensemble algorithm combine them.

question b: neural network, or other similar model not work due dimensionality. amazon.com , want input item set of 5 items. might encode input neuron count equal 1 neuron each item times 5. how many items on amazon. have no idea, guessing on 100k. if creative dimensionality reduction, going massive neural network, or back upwards vector machine or random forest.

algorithm neural-network