Designing and Understanding Profit-Maximizing Recommender Systems for Online Retailing (Acronym: DUPSOR)

• Applicant: Professor Dr. Oliver Hinz
• Subject Area: Operations Management and Computer Science for Business Administration
• Term: since 2020
• Project identifier: Deutsche Forschungsgemeinschaft (DFG) - Project number 449023539

The overall goal of our project is to design and understand profit-maximizing recommender systems (PMRS) in e-commerce. In the initial phase of our project we have already achieved the important goals and designed, tested, and analyzed the impact of PMRS. The potential continuation phase of the project has three (new) major goals. The first objective is to generalize and extend the findings on effective PMRS design and impact from the initial phase. Indeed, the literature suggests that both the type of goods and the deployment duration moderate the effects of PMRS on customers, sales, and profits. To investigate this, we extend our field experiments from hedonic goods (here: games) to utilitarian goods (here: office supplies). We achieve this through a new cooperation with an online retailer of office supplies that generates €315 million in annual sales and has already agreed to the corresponding field experiments. In addition, the duration of the current field experiment will be extended beyond six months to capture long-term effects. The second objective is to analyze the role of explanations in PMRS (i.e., explained recommendations) in a real-world setting. Prior research has mainly looked at explanations in rule-based systems in controlled laboratory settings. We build on our own findings from laboratory environments, on literature, and on new field experiments to analyze the effect of explanations on trust and acceptance of recommendations in e-commerce. Special attention is given to the application of AI and explainable AI in recommendation systems in online retail. The third objective is to investigate the integration of Large Language Models (LLMs) such as ChatGPT into recommender systems (and PMRS in particular) in e-commerce. The adoption of LLMs is growing rapidly in a variety of domains. Integrating LLMs potentially enables automated and interactive recommendations at scale, and thus could improve the customer experience. At the same time, it poses risks, such as privacy breaches and biased recommendations. We aim to develop frameworks for the responsible and purposeful use of LLMs in recommender systems (and PMRS in particular), and to assess their impact on customer behavior and satisfaction, as well as revenue and profit.

• DFG Programme: Research Grants
• International Connection: Israel
• International Co-Applicant: Professor Dr. Shachar Reichman