We wish you a good start into the new year 2021! Perhaps you have been inspired by automatic product recommendations in the last few months and perhaps you have discovered new products that you have subsequently purchased. We also encounter automatic product recommendations in the field of digital television, e.g. at Netflix or Amazon Prime Video. Such systems use machine learning to find out which movies and series you are most likely to like. Unfortunately, even the best recommendation algorithms currently available are mistaken in more than 2 out of 10 cases, which means that we regularly receive recommendations that we find unhelpful. One problem with such recommendations is that most common systems do not provide convincing explanations why the algorithm has chosen a certain product for you as a consumer. Sometimes you read that a product is recommended based on your browsing history, and sometimes it is even products that you have bought long ago and do not want to buy a second time.
The marketing scholars Professor André Marchand and Jun. Prof. Paul Marx (University of Siegen) have now developed a new algorithm that addresses exactly this problem. The algorithm uses a combination of content-based and collaborative filtering. The empirical experiments with two real data sets with more than 100 million product ratings show that this new algorithm exceeds the established recommendation approaches in terms of prediction accuracy (more than 5% better than the Netflix award winning algorithm) and its ability to provide actionable explanations, which is also an ethical requirement for artificial intelligence systems.
This new study has been published in the Journal of Retailing and can be downloaded here!