gMPIS: maximal-profit item selection based on generalized cross-selling considerations

An algorithm was proposed to rank items with respect to profits and to select the most profitable ones for business and other applications by analyzing historical transaction data sets. The decision of maximal-profit items subset considers both the cross-selling effects among items and the customers' changing buying behaviors. Generalized loss rules are proposed to model the effects of the unselected items to the customers' purchase actions. We show that the approach models the customers' buying behavior well and is very suitable and practicable for real business applications. We propose a fast heuristic algorithm to the problem. Experiments show that the algorithm is highly effective and efficient and is scalable for large data sets.