Abstract
The following paper presents our work in progress to deal with time dependence on e-commerce recommending system.We are particulary interested in rather specific e-commerce sub-domain with low consumption rate and not very loyal customers. Time dependence in such domains is rather lower than e.g. in multimedia portals, however our experiments corroborated that it still plays an important role in quality of recommendations.
Experimented methods shown decreasing recommendation quality (according to nDCG and average position) with increased time distance, however the overally best method, content boosted matrix factorization suffers from more significant decline. In conclusion, we point out several approaches how to further improve recommendations by incorporating time-awareness.