Predicting Human Behavior Using User’s Contextual Embedding by Convolution of Action Graph

Abstract

Recently, predicting human behavior using logs that include user location information and categories of facilities visited has been actively researched. However, not enough research focused on user behavioral embedding expressing user preferences. In this research, we build an action graph with categories as nodes and transitions between categories as edges in order to capture the transitions of preference in consideration of the context of the places visited by users. Then, we propose a behavior prediction model that uses features of action graph extracted by the graph convolutional networks. In experiments, we present that proposed model using user embedding extracted by graph convolution are improving accuracy.