Context-aware recommendations

[psychogyiokostas]

Professor @hwwang55 thank you for your work and for providing us the code for RippleNet.
RippleNet as far as i'm concerned does not take into consideration the timing and the order in which a user has interacted with certain items. For instance a user may have liked items A, B and C ordered by time, from older to more recent selections. We would expect that item C would contribute more than B, and B would contribute more than A in the resulting recommended items.

In that sense, could we modify RippleNet to be context-aware? For instance, add a lower weight and down-rank seed items' embeddings that were selected least recently than others?

Thank you for your time, looking forward for your reply.