I'm currently exploring articles on Neo4j recommendation systems. Here's the problem:
- in a stock replenishment model for any given item there are at least five to six substitute items;
- there can also be a case where one "bundled" item can be replaced with several individual parts to get the initial "bundled" item;
- the problem persist when the inventory/stock needs to be replenished - which substitute item should be ordered for the inventory if there is no initial (original or preferred) item available;
- if i am not mistaken, this is in fact a problem of a recommendation system, but not for the customers - rather category managers and similar roles;
- if so, i need to start sketching a new model, but need answers on the questions written below...
Where to start?
- are there any whitepapers discussing similar problems?
- my idea is to make a random .csv file with 100 original items, each having 5-10 substitute items;
- how many attributes for nodes/items should be initially considered? For example - brand, price, category listing (premium, average, budget, etc.)...?
- what about links?
- what kind of calculations should be used within GDS to get some kind of a meaningful output? I'm doubting centrality measures could help here.