krishnacharya / recsys_eq

Code for Producers Equilibria and Dynamics in Engagement-Driven Recommender Systems

recsys_eq

Producers Equilibria and Dynamics in Engagement-Driven Recommender Systems

Abstract

Online platforms such as YouTube, Instagram, TikTok heavily rely on recommender systems to decide what content to show to which users. Content producers often aim to produce material that is likely to be shown to users and lead them to engage with said producer. To do so, producers try to align their content with the preferences of their targeted user base. In this work, we explore the equilibrium behavior of producers that are interested in maximizing user engagement. We study two variants of the content-serving rule that the platform's recommender system uses, and we show structural results on producers' production at equilibrium. We leverage these structural results to show that, in simple settings, we see specialization naturally arising from the competition among producers trying to maximize user engagement. We provide a heuristic for computing equilibria of our engagement game, and evaluate it experimentally. We show i) the performance and convergence of our heuristic, ii) the producer and user utilities at equilibrium, and iii) the level of producer specialization.

Organization:

• The core classes and helper functions are in the source directory.
• The ProducersEngagementGame in Producers.py instanciates an Engagement Game and requires the number of producers, the content serving rule (softmax or linear), and a user object(see Users.py).
• The workflow involves creating an ProducersEngagementGame object, then calling best_response_dynamics on it, all the scripts in the scripts directory follow this workflow.
• We use the following values for the random seeds = [2, 3, 5, 7, 11, 13, 17, 19, 23, 29], dimension of embeddings = [5, 10, 15, 20] and the number of producers = np.arange(10, 110, 10).
• The results for the number of iterations till convergence to Nash equilibrium and utilities at NE are available in the folders csv_results/br_dynamics and csv_results/utility-tables respectively.
• The pandas Dataframes for each of the 400 instances (of seed x dimensions x producers) for all the experiments are available in the saved_frames folder.
• All the figures in the paper are in the plots folder.

Conda environment is available in recsys_eq.yml