Graphs for recommendation systems

This is an older note:

I think using graphs for recommendation systems is not a natural choice where temporal classification and sequences may play some role. It is not natural to represent time with nodes where each can have multiple edges. Instead RNN or Markov chains might be more suitable. Recommendation is almost always sequential. Otherwise it’s for finding embeddings between elements. If then graph is not fitting to these problems. If graphs are not suitable, why do we use them?

The answer is simple: People don’t buy things in sequence, they have options and multiple choices and factors that play role in these choices. So it’s actually natural to model these with graphs.