Companies might be inclined to recommend products with the highest profit rating or unsellable stock using recommendation agents. This paper presents and tests countermeasures for biased product recommendation agents. The researchers conducted an experiment in which they exposed users of an e-Commerce website with warning messages about the risks of relying on recommendations provided by the website in three different styles; a warning without advice, a warning with positively framed advice and a warning with negatively framed advice.
All three of the methods led to an increased perceived bias by the users of the website in comparison to the control group, indicating that showing a warning message influences the choices a consumer makes after being warned.
The authors conclude that the best method to warn consumers about biased product recommendations is the negatively worded warning. This method led to the highest score of perceived bias by the users, and was the only method that did not increase the number of false positives reported by the users.
The use of recommendation agents to recommend the most profitable product is a valid concern and widely discussed and proven in other research, providing a good practical example to start the research.
The authors have countered the biased recommendations using simple measures that could easily be implemented in the real world.
In the supplementary research the authors note that the number of users using the search-by-brand functionality to verify the suggestions made by the product recommendation system increased from 12.9% to 43.1%. This suggests that the users did not simple suggest bias, but actively used the tools at their disposal to verify the validity of the warning message.
By using the perceived bias as the DV, the authors measured the opinion of the user whether they thought they were being tricked or not. This might have left out valuable results of people who were tricked but were not aware of it even after receiving the warning. Since the study contained relatively few choices, the authors could also have used the actual value for money rating of the selected products. This would have provided a more accurate view of whether people were influenced by the biased recommendation system.
The proposed solution of a browser extension feels a bit too extensive. How many times do you need to be reminded to be wary of bias in product recommendations? A government issued cautionary commercial (SIRE in the Netherlands) could possibly achieve the same results by triggering people to think before trusting a recommendation.