A comparison of learning schemes for recommender software agents: a case study in home furniture

摘要

Recommender agents will personalise the shopping experience of e-commerce users. In addition, the same technology can be used to support experimentation so that companies can implement systematic market learning methodologies. This paper presents a comparison regarding the relative predictive performance of Backpropagation neural networks, Fuzzy ARTMAP neural networks and Support Vector Machines in implementing recommendation systems based on individual models for electronic commerce. The results show that support vector machines perform better when the training data set is very limited in size. However, supervised neural networks based on minimising errors (i.e., Backpropagation) are able to provide good answers when the training data sets are of a relatively larger size. In addition, supervised neural networks based on forecasting by analogy (i.e., Fuzzy ARTMAP) are also able to exhibit good performance when ensemble schemes are used.