Time-aware recommender systems (TARS) are systems that take into account a time factor - the age of the user data. There are three approaches for using a time factor: (1) the user data may be given different weights by their age, (2) it may be treated as a step in a biological process and (3) it may be compared in different time frames to find a significant pattern. This research deals with the latter approach.

When dividing the data into several time frames, matching users becomes more difficult - similarity between users that was once identified in the total time frame may disappear when trying to match between them in smaller time frames.

The user matching problem is largely affected by the sparsity problem, which is well known in the recommender system literature. Sparsity occurs where the actual interactions between users and data items is much smaller in comparison to the entire collection of possible interactions. The sparsity grows as the data is split into several time frames for comparison. As sparsity grows, matching similar users in different time frames becomes harder, increasing the need for finding relevant neighboring users.

Our research suggests a flexible solution for dealing with the similarity limitation of current methods. To overcome the similarity problem, we suggest dividing items into multiple features. Using these features we extract several user interests, which can be compared among users. This comparison results in more user matches than in current TARS.