The case for graph-based recommendations

Recommender systems have been intensively used to create personalised profiles, which enhance the user experience. In certain areas, such as e-learning, this approach is short-sighted, since each student masters each concept through different means. The progress from one concept to the next, or from one lesson to another, does not necessarily follow a fixed pattern. Given these settings, we can no longer use simple structures (vectors, strings, etc.) to represent each user's interactions with the system, because the sequence of events and their mapping to user's intentions, build up into more complex synergies. As a consequence, we propose a graph-based interpretation of the problem and identify the challenges behind (a) using graphs to model the users' journeys and hence as the input to the recommender system, and (b) producing recommendations in the form of graphs of actions to be taken.