Using natural language processing analysis to investigate daily-life positive autobiographical memories

Abstract

Recalling positive autobiographical memories (AMs) can improve wellbeing, yet few studies capture these memories as they naturally occur or examine their detailed content. In a sample of 70 community adults, we used ecological momentary assessment to assess naturally occurring positive AMs in daily life. Using the large language model-based Python library BERTopic, we identified eight underlying topics: ‘Friends, partner, family’ (68.42%), ‘Work, study’ (8.33%), ‘Being in nature or the moment, pets’ (5.60%), ‘Family & social events’ (3.69%), ‘Exam success’ (3.96%), ‘Success in sports – self or as a fan’ (3.28%), ‘Outdoor activities and sports’ (3.14%), and ‘School days, studies’ (2.19%). Participants reported vivid, sensory AMs accompanied by positive emotions. Greater AM vividness was systematically associated with greater immediate AM-related happiness, underscoring vividness as a potential amplifier in everyday remembering. By sharing our detailed methodology, code, and anonymized examples, we lay the groundwork for large-scale computational analysis of AMs assessed under the noisy and unstructured conditions of everyday life.