Decoupled Recommender Systems: Exploring Alternative Recommender Ecosystem Designs

TL;DR

This paper addresses whether decoupling recommender algorithms from the platforms they serve can improve outcomes for niche consumers and providers. It introduces SMORES, a simulation framework that models interactions among consumers, providers, and multiple recommender platforms, and evaluates two switching strategies (threshold-based and UCB) across a mainstream and a niche recommender. Using a MovieLens-based dataset with manipulated genre preferences, the study finds that a decoupled, algorithm-market-like ecosystem improves niche utility and provides a pathway for niche providers to prosper, while mainstream participants may experience reduced exposure. The work demonstrates the potential of decoupled architectures to diversify the recommender ecosystem and motivates further exploration of multiple algorithms, contexts, and business models in practical deployments.

Abstract

Recommender ecosystems are an emerging subject of research. Such research examines how the characteristics of algorithms, recommendation consumers, and item providers influence system dynamics and long-term outcomes. One architectural possibility that has not yet been widely explored in this line of research is the consequences of a configuration in which recommendation algorithms are decoupled from the platforms they serve. This is sometimes called "the friendly neighborhood algorithm store" or "middleware" model. We are particularly interested in how such architectures might offer a range of different distributions of utility across consumers, providers, and recommendation platforms. In this paper, we create a model of a recommendation ecosystem that incorporates algorithm choice and examine the outcomes of such a design.

Figures (6)

• Figure 1: The monolithic vs decoupled recommender system architecture. In the decoupled scenario, consumers can choose among recommender systems and both consumers and providers benefit.
• Figure 2: The simulation environment process highlighting the stakeholders in green, processes in blue, and the inputs/outputs in pink.
• Figure 3: Single recommender experiment: Consumer and provider utility, distribution and cumulative results. The top left box plot shows consumer utility on the last day, while the bottom left strip plot shows provider utility on the last day. The right column line plots depict cumulative consumer utility and provider utility over all cycles.
• Figure 4: Decoupled recommender with threshold-based switching experiment. The dashed line in the top left plot represents the switching threshold. The provider utility is the sum of the two recommenders' utility for that provider.
• Figure 5: Decoupled recommender with UCB-based switching.