AI驅動之同伴支持媒合系統：促進線上心理健康社區之認知再評估

認知再評估（reappraisal）是一項有助人類負面情緒與想法調節的關鍵策略，不過，當面對具有壓力的狀況時，要自行靈活應用、實現認知再評估通常並不容易，因此，線上心理健康社區（online mental health communities，OMHCs）成為了新興趨勢，供人們尋求及提供幫助。然而，現存研究發現， OMHCs提供支持觀點的內容品質低落，大多採用事後才篩除不適言論或事前訓練支持提供者（support provider）來解決此問題，此兩種解決方案皆存在效率不彰的問題，也會拉低平台使用的留存率。本研究提出AI驅動心理健康互助平台：VoissBot，搭載媒合參考答案之功能，在使用者提供互助內容後即時分析用戶語意並給予適配的參考答案，用以達到長期、潛在提升支持提供者幫助他人進行認知再評估的能力。正式實驗中，受測者分為A、B組，分別媒合相似與歧異觀點，透過比較兩組的支持品質，得出較好的系統媒合機制。我們邀請四位心理專家訂定準則，首先對實驗前置準備數據進行人工標記，並定義出關鍵七大分類標籤，接著本研究基於此結果訓練出認知再評估多標籤分類器，準確率達89%。正式實驗階段，本研究套用上述分類器來比較A、B兩組結果後，發現B組給予歧異觀點的自動化媒合機制，相較於A組，更能在無須矯正支持提供者觀點的條件下潛在提升其支持品質（p<0.001），結合心理師專家歸納的觀點，我們進一步探究如何建置能真正促進用戶認知再評估的AI驅動心理健康互助平台。

關鍵字

自然語言處理；人工智慧媒合系統；心理健康；深度學習；社群網路

並列摘要

Reappraisal is an essential technique for people to transform their construal of negative events. To facilitate reappraisal, many people have turned to online mental health communities (OMHCs) to seek support; however, prior research found that the support quality of contents provided in OMHCs is unstable. This paper proposes an AI-driven system, VoissBot, equipped with a matching function to immediately analyze support contents' semantics and output a corresponding matching reply to the users. We conducted a 10-day experiment with 155 participants and separated them into two groups to complete reappraisal supporting tasks with corresponding matching or dissimilar matching algorithms. In addition, four licensed therapists were recruited to define different levels of support quality from the participants' comments and yield seven labels regarding support quality. Finally, we trained a multi-label classifier for cognitive reappraisal based on the classification and their hand-coded data, and the classifier accuracy rate reached 89%. Our results show dissimilar matching replies elicit a higher level of empathy (p<0.001), and the users need minor corrections to improve their support quality. Thus, our study demonstrates a promising way to build an AI-driven system that facilitates user reappraisal.

並列關鍵字

Natural Language Processing ； Artificial Intelligence Matching System ； Mental Health ； Deep Learning ； Social Networks

參考文獻

1. Denny, B.T. and K.N. Ochsner, Behavioral effects of longitudinal training in cognitive reappraisal. Emotion, 2014. 14（2）: p. 425-433.

Google Scholar

2. Zhang, C.Q., et al., Occupational stressors, mental health, and sleep difficulty among nurses during the COVID-19 pandemic: The mediating roles of cognitive fusion and cognitive reappraisal. J Contextual Behav Sci, 2021. 19: p. 64-71.

Google Scholar

3. Blechert, J., et al., See What You Think:Reappraisal Modulates Behavioral and Neural Responses to Social Stimuli. Psychological Science, 2012. 23（4）: p. 346-353.

Google Scholar

4. Doré, B.P., et al., Helping Others Regulate Emotion Predicts Increased Regulation of One's Own Emotions and Decreased Symptoms of Depression. Personality amp; social psychology bulletin, 2017. 43（5）: p. 729-739.

Google Scholar

5. Morris, R. and R. Picard, Crowdsourcing Collective Emotional Intelligence. 2012.

Google Scholar

國際替代計量

AI驅動之同伴支持媒合系統：促進線上心理健康社區之認知再評估

全文下載

主題瀏覽