System-2 Recommenders: Disentangling Utility and Engagement in Recommendation Systems via Temporal Point-Processes

Recommender systems are an important part of the modern human experience whose influence ranges from the food we eat to the news we read. Yet, there is still debate as to what extent online recommendation platforms are aligned with the goals of their users. A core issue fueling this debate is the challenge of inferring a user's utility based on their engagement signals such as likes, shares, watch time etc., which are often the primary metric used by platforms to optimize content. This is because users' utility-driven decision-processes (which we refer to as System-2), e.g., reading news that are accurate and relevant for them, are often confounded by their impulsive or unconscious decision-processes (which we refer to as System-1), e.g., spend time on click-bait news articles. As a result, it is difficult to infer whether an observed engagement is utility-driven or impulse-driven. In this paper we explore a new approach to recommender systems where we infer user's utility based on their return probability to the platform rather than engagement signals. This approach is based on the intuition that users tend to return to a platform in the long run if it creates utility for them, while pure engagement-driven interactions, i.e., interactions that do not add meaningful utility, may affect user return in the short term but will not have a lasting effect. For this purpose, we propose a generative model in which past content interactions impact the arrival rates of users based on a self-exciting Hawkes process. These arrival rates to the platform are a combination of both System-1 and System-2 decision processes. The System-2 arrival intensity depends on the utility drawn from past content interactions and has a long lasting effect on return probability. In contrast, System-1 arrival intensity depends on the instantaneous gratification or moreishness and tends to vanish rapidly in time. We show analytically that given samples from this model it is provably possible to disentangle the System-1 and System-2 decision-processes and thus infer user's utility, thereby allowing us to optimize content based on it. We conduct experiments on synthetic data to demonstrate the effectiveness of our approach over engagement optimization.