Our study aims to develop an engaging interaction system that assists dementia patients who have frequent memory loss. The system features include: videoing patients' daily social encounters, then replaying their key events at arranged or appropriate times, which not only functions to give patients reminders, but also motivates human machine interaction through multimodal digital memory hooks. In the longer term we will explore how developing a novel human-machine interaction can lead to defining a nonpharmacological intervention for patients with dementia. In the early stage of development, we conducted interviews with care staff and clinical professionals (n=17) to identify what kinds of interactions and conversations are most significant and/or most useful for the development of the memory assistant that has the potential both for refreshing of memories and driving them into long term memory. We have further investigated the semiautomated identification of relevant scenes from video recordings of daily interactions. The analysis aims to identify key target memories of patients that either represent significant family/life events or enrichen the digital memory hooks necessary for such events to be retained in long term memory.