Coping With Imbalanced Data in the Automated Detection of Reminiscence From Everyday Life Conversations of Older Adults

Reminiscence—the act of recalling or telling others about relevant personal past experiences—plays an important role in the well-being older adults. Therefore, it is to develop intelligent systems aiming at improving elderly by reliably detecting reminiscence their everyday life conversations. Data imbalance one main challenges automatic detection from conversations, as reminiscing a rare event. In this paper, we address problem proposing methodology for coping with imbalanced data conversations The combines augmentation using BERT (Bidirectional Encoder Representations Transformer) and feature extraction techniques leveraging natural language processing German language. We evaluate proposed on dataset comprising transcripts social adults held German. compare our results previous work addressing same show that approach strongly outperforms baseline. study may support development real-time design digital health interventions well-being.