Quadruplet Alignment Loss Function for Representing the Stages of Behavior Change

Abstract

This study proposes a method to estimate the stages of behavior change in research activities using text data from progress reports submitted by students periodically. Motivation is crucial in student research activities. It is essential for the supervisor to provide students with guidance tailored to their motivational states. However, since motivation levels vary with students, uniform instruction fails to yield sufficient results. To provide appropriate support, it is necessary to understand the motivation level of each student. Existing methods for measuring motivation tend to rely on subjective evaluations, placing a significant burden on respondents. The method focuses on behavioral changes reflected in progress reports periodically presented. The text data from progress reports is converted into vectors indicating the stage of behavior change through the Quadruplet Alignment Loss proposed in this study. The loss function models the sequential relationship among the contemplation stage, the preparation stage, and the action stage. The results of the experiment have confirmed that the proposed method improves the estimation accuracy of the stage of behavior change. In particular, the estimation accuracy during the preparation stage has significantly improved, which demonstrates successful estimation of the critical transitional stage leading to action. It suggests the potential to support supervisors in providing research guidance tailored to students’ individual motivation.