I would like to develop a recommendation system (probably hybrid, user-based and feature-based) for items which are organized in a tree (there are categories, divided in sub-categories, divided in sub-sub-categories). The branches have different depths. An item is associated to a unique node of this tree, and it can be at any level. The tree has about 100 nodes, and it is probably the most important information I have about my items.
Here is what I found so far :
- This is a very classic situations (any shop organizes its products in categories like "electronics"/"cloths"/... and in sub-categories like "phones"/"PC"/...) but I was surprised to find very limited information about it (but I maybe used non-optimal keywords).
- There is some work about graphs (which may be a term similar to "tree"), but it seems to be a different problem (e.g. graphs of people in contact with each other like in Facebook)
- Here is a 4 year old message with an answer about matrix factorisation which didn't really help me (I maybe didn't understand it) : Which algorithms should I use for recommendation system using a graph database?
- Here is a publication by Alibaba where the tree seems to be only used to "break down" the problem in smaller pieces : https://medium.com/@alitech_2017/improving-recommender-system-with-tree-based-deep-model-e69f4a32d014
Would you have any advice ?