Method of determining user preferences for the personalized recommender systems for public transport passengers

Abstract

The question of creating a navigation recommender system based on user preferences arose with the development of recommender systems. The paper presents the theoretical and algorithmic aspects of making a personalized recommender system (mobile service) designed for public transport users. The main focus is to identify and formalize the concept of “user preferences”, which is based on modern personalized recommender systems. Informal (verbal) and formal (mathematical) formulations of the corresponding problems of determining “user preferences” in a specific spatial-temporal context are presented: the preferred stops definition and the preferred “transport correspondences” definition. The first task can be represented as a classification problem. Thus, it represented using well-known pattern recognition and machine learning methods. In this paper, we use an approach based on the estimation algorithm proposed by Yu.I. Zhuravlev and nonparametric estimation of Parzen probability density. The second task is to find estimates for a series of conditional distributions. The experiments were conducted on data from the mobile application “Pribyvalka-63”. The application is a part of the tosamara.ru service, currently used to inform Samara residents about the public transport movement.