Estimating Adaptive Individual Interests and Needs Based on Online Local Variational Inference for a Logistic Regression Mixture Model

In real companies engaged in economic activities through transactions involving consumer items, such as retail, distribution, finance, and information materials, supplying an opportunity to customers to choose specialized items is an important factor that can improve customer satisfaction and convenience allowing their diverse and time-dependent needs to be met. However, capturing the specialized needs of customers accurately is a difficult task because their needs depend on time, context, situation, and meaning. Recently, physical computational environments have been developing rapidly, thereby allowing easy implementation to sense a customer's action and deal with it sequentially. In this paper, we propose a personalized method to predict individual interests and demands appropriately. In particular, the system learns the customers' situation, meaning, and action from their action history, and reflects a feedback of the result to predict the next action. To realize this method, we utilize the following two methodologies: the mathematical model of meaning (MMM), which is a semantic associative search technology; and the local variational inference (LVI), which is an approximation of the Bayesian inference. A numerical experiment shows that the proposed method performed better than a typical method.