ARE WE MAKING APPROPRIATE RECOMMENDATIONS?-A PERSPECTIVE FROM MULTIDIMENSIONAL ITEMRESPONSE THEORY ON 'INTERNET RECOMMENDATIONSYSTEMS'

摘要

This project describes a new model to design Internet recommendation systems based onMultidimensional Item Response Theory (MIRT). The MIRT model allows statistical integration offive types of information useful for making recommendations: a person's expressed preferences,preferences of other consumers, expert evaluations, item characteristics, and individualcharacteristics. In addition, the effect of unobserved product heterogeneity on consumerpreferences is introduced in the model.