An automated approach to estimate human interest

摘要

Can we model and estimate interest? In general, when an individual engages with an object, say Facebook, Instagram, a Mobile game, or anything else, we know that there is some interest that the person has in the object. However, we do not have a procedure that can tell us by “how much” of a factor is the person interested. Simply put, can we find a “number” for someone’s interest? In this article, we propose the design of a framework that can handle this issue. We formulate the interest estimation problem as a state estimation problem and deduce interest indirectly from the activity. Activity, stimulated by interest, is measured via a subjective-objective weighted approach. Further, we present a novel continuous-time model for interest by drawing inspiration from Physics and Economics simultaneously. We model interest along the Ornstein-Uhlenbeck process in Physics and improve the performance by borrowing ideas from Stochastic Volatility Models in Economics. Subsequently, we employ particle filter to solve the interest estimation problem. To validate the feasibility of the proposed theory in practice, we investigate the model by conducting numerical simulations on real-world datasets. The results demonstrate good performance of the framework, and thus match the theoretical expectations from the method. Lastly, we implement the framework in practice and deploy it as a RESTful service, thereby providing a uniform interface for accessing the procedure via any remote or local application.