Present Bias Estimation using Behavior History

Abstract

Present bias is the bias to be overconfident of the patience in the distant future and impatient in the near future. People with a strong present bias tend to procrastinate their tasks and are vulnerable to intermediate temptations, and people often fail to achieve long-term goals due to present bias. In the result, people with a strong present bias develop bad lifestyle habits such as smoking or obesity. Intervention applications considering present bias in individuals may improve their lifestyle habits. In this paper, we propose a present-bias estimation method using history of daily behaviors collected by wearable devices. The proposed method employs Transformer that shows high performance in processing sequential data. We propose new positional encoding method considering timestamp and event occurrence information for accurately processing behavior histories. Our proposal showed higher estimation results compared with other positional encodings.