Feedback Loops in Recommender Systems

As we progress through this course, one thing that we do know for sure us that recommendation systems are designed to help people make decisions. These systems are commonly used on online platforms where data is always updated. They influence how users perceive the world around them by filtering access to media such as political opinions or opinions that differ from one’s point of view. As a result of all the updates, a feedback loop is created.

In the world today where technology is constantly emerging, machine learning is being extensively used in recommender systems and so the decisions made by these systems can influence users’ beliefs and preferences. This in turn affect the feedback that the learning system receives hence creating a feedback loop. For instance, live systems are updated or retrained regularly to incorporate new data that was influenced by the recommendation system itself, forming a feedback loop. In this case issues of fairness arise because let’s say there’s a distinct group of minorities who have a certain preference. The system might undermine the needs of that group to optimize utility for the majority.

Furthermore, these systems impact critical decision-making techniques including but not limited to:

It is with hope that researchers and practitioners will:

  1. Actively assess systems with –

    • Diversity
    • Serendipity
    • Novelty
    • Coverage

  2. Evaluate the distribution that may impact all users instead of exclusively reporting averages.

After all, these systems were developed by human and are contantly upgraded to be like us, only better and faster. Therefore some human bias must exist in recommender systems. However, the systems are learning and may help mitigate human bias inorder to prevent unethical targeting or customer segmentation.