Complex splitting of context-aware recommendations

Item splitting is an effective approach to improve the prediction accuracy of contextual recommendations. In this approach, an item is split into two items under two alternative contextual conditions, respectively. In this work, complex splitting is proposed to get more specialised rating data and further improve the accuracy of the recommendations. The key to the approach is to select multiple contextual conditions for splitting the user or item. We translate this into a contextual conditions combinatorial optimisation problem based on a discrete binary particle swarm optimisation algorithm. The item or user is split into two different items or users according to those contextual conditions in optimal combination: one is rated in a context that meets all the contextual conditions of the best combination, and the other one is rated in a context that does not. In this way, more specialised rating data can be obtained, which results in a more accurate recommendation when the data is input into the recommend...