本研究旨在探討如何將大語言模型(LLM)應用於解決人口老齡化帶來的居家照護資源短缺問題。我們透過主流大語言模型及提示詞工程，評估LLM在理解並處理各種居家照護情境主動式照護的能力。主要實驗包括角色預測、情感分析、角色情感回應和聲音事件分析。
　　在角色預測能力方面，LLM表現出平均99.16%的高準確率，顯示LLM有潛力從對話中快速識別被照護者的個性特徵與語言風格，從而提供個性化的照護服務。情感分析方面，採用Ekman的6個情感分類法，LLM較易達成多數決，可以簡化應對複雜度並提高回應速度，滿足照護情境需求。角色情感回應測試中，LLM平均達到78.6%的角色預測準確率，展示了其模擬不同角色言語風格的能力。聲音事件反應測試結果則顯示，LLM能夠合理分析聲音事件，並提供適當的應對策略，展現了在緊急情況下的決策能力。
　　未來結合多模態數據（如語音、表情）進行情感分析，有望比純文字分析提升辨識準確度，能更全面地理解被照護者的狀態。生成文化偏差問題突顯了發展本土化AI模型的必要性。在特定應用下限制模型生成範圍，都是未來研究需要關注的方向。總的來說，本研究為LLM在老年照護領域提供了有價值的洞察。通過進一步的研究和優化，LLM有望在提升照護質量、緩解人力資源壓力上發揮作用，為應對人口老齡化帶來的挑戰開闢新的可能性。這項研究不僅為智能照護系統的發展提供了重要基礎，也為未來更人性化、高效的老年照護模式指引了方向。

This study aims to explore how Large Language Models (LLMs) can be applied to address the shortage of home care resources resulting from an aging population. Through generative large language models and prompt engineering, we evaluate the ability of LLMs to understand and process various home care scenarios for proactive care. The main experiments include role prediction, sentiment analysis, role-based emotional response, and sound event analysis.
In terms of role prediction capability, LLMs demonstrated a high average accuracy of 99.16%, indicating their potential to quickly identify the personality traits and language styles of care recipients from conversations, thereby providing personalized care services. Regarding sentiment analysis, using Ekman's six emotion classification method, LLMs more easily reached a majority decision, which can simplify response complexity and increase response speed, meeting the needs of care scenarios. In the role-based emotional response test, LLMs achieved an average role prediction accuracy of 78.6%, showcasing their ability to simulate different role-specific speech styles. The results of the sound event reaction test indicated that LLMs can reasonably analyze sound events and provide appropriate response strategies, demonstrating decision-making capabilities in emergency situations.
Future integration of multimodal data (such as voice and facial expressions) for sentiment analysis is expected to improve recognition accuracy compared to pure text analysis, enabling a more comprehensive understanding of the care recipient's state. The issue of generated cultural bias highlights the necessity of developing localized AI models. Restricting the model's generation range for specific applications is also a direction that future research needs to address. Overall, this study provides valuable insights into the application of LLMs in elderly care. Through further research and optimization, LLMs have the potential to play a role in improving care quality and alleviating human resource pressure, opening up new possibilities for addressing the challenges posed by an aging population. This research not only provides an important foundation for the development of intelligent care systems but also points the way toward more humane and efficient elderly care models in the future.

[1] 強化人口及移民政策高齡化說明. (2022). 中華民國國家發展委員會. https://www.ndc.gov.tw/Content_List.aspx?n=2688C8F5935982DC
[2] Mikolov, T., et al. (2013). Efficient estimation of word representation in vector space. arXiv:1301.3781
[3] Dupond, D. (2019). A thorough review of the current advance of neural network structures. Annual Reviews in Control. 14, 200-230.
[4] Vaswani, A., et al. (2017). Attention is all you need. arXiv:1706.03762.
[5] “Better Language Models and Their Implications”. OpenAI (2019). https://openai.com/index/better-language-models/. Accessed 26 July 2024.
[6] Bowman, S.R. (2023). Eight things to know about large language models. arXiv:2304.00612.
[7] Prompt engineering. OpenAI. https://platform.openai.com/docs/guides/prompt-engineering. Accessed 26 July 2024.
[8] 葉淑娟、施智婷、莊智薰、蔡淑鳳. (2004). 社會支持系統與老年人生活滿意度之關係－以高雄市老年人為例. 中山管理評論, 12(2), 399-427. https://doi.org/10.6160/2004.06.06
[9] 陳淑汝（2015）。遷居安養機構老人晚年生命經驗之探討。﹝碩士論文。國立暨南國際大學﹞臺灣博碩士論文知識加值系統。 https://hdl.handle.net/11296/yn6934。
[10] 吳俊賓（2020）。聊天機器人對話式介面設計以老人照護為例。﹝碩士論文。淡江大學﹞臺灣博碩士論文知識加值系統。 https://hdl.handle.net/11296/eg65dx。
[11] Brown, T. B., Mann, B., Ryder, N., Subbiah, M., Kaplan, J., Dhariwal, P., Neelakantan, A., Shyam, P., Sastry, G., Askell, A., Agarwal, S., Herbert-Voss, A., Krueger, G., Henighan, T., Child, R., Ramesh, A., Ziegler, D. M., Wu, J., Winter, C., Hesse, C., Chen, M., Sigler, E., Litwin, M., Gray, S., Chess, B., Clark, J., Berner, C., McCandlish, S., Radford, A., Sutskever, I., Amodei, D. (2020). Language models are few-shot learners. arXiv. https://doi.org/10.48550/arXiv.2005.14165
[12] Padhan, S., Mohapatra, A., Ramasamy, S. K., & Agrawal, S. (2023). Artificial intelligence (AI) and robotics in elderly healthcare: Enabling independence and quality of life. Cureus, 15(8), Article e42905. https://doi.org/10.7759/cureus.42905
[13] 陳愷昀. (2023, May 4). 【地球圖書館】「我可以照顧你嗎？」解放護理人員　AI機器人要取代人類還是打輔助？. 地球圖輯隊. https://dq.yam.com/post/15375. Accessed 26 July 2024.
[14] Bates, J. (1994). The role of emotion in believable agents. Communications of the ACM, 37(7), 122-125. https://doi.org/10.1145/176789.176803
[15] Park, J. S., O'Brien, J., Cai, C. J., Morris, M. R., Liang, P., & Bernstein, M. S. (2023). Generative agents: Interactive simulacra of human behavior. arXiv:2304.03442
[16] NHS England. (n.d.). Proactive care: Providing care and support for people living at home with moderate or severe frailty. https://www.england.nhs.uk/long-read/proactive-care-providing-care-and-support-for-people-living-at-home-with-moderate-or-severe-frailty/. Accessed 26 July 2024.
[17] I Ran Tests — ChatGPT 4 vs Claude 3 Sonnet, Who Wins? (2024, May 6). Medium. https://generativeai.pub/i-tested-out-chatgpt-4-vs-claude-3-sonnet-who-wins-df7aba30d373. Accessed 26 July 2024.
[18] Wotakugo. (2023, September 19). 52+ Classic Howl’s Moving Castle Quotes That Bring Back Memories. Medium. https://medium.com/@contact_72457/52-classic-howls-moving-castle-quotes-that-bring-back-memories-56a44081330f
[19] Quotes from Mother Teresa. (n.d.). TurnBull High School. https://blogs.glowscotland.org.uk/ed/turnbullvva/quotes-from-mother-teresa/. Accessed 26 July 2024.
[20] Good-Will-Hunting-Entire-Screenplay. (2012, April). Ivana Chubbuck Studio. https://www.ivanachubbuck.com/wp-content/uploads/2012/02/Good-Will-Hunting-Entire-Screenplay.pdf. Accessed 26 July 2024.
[21] Liles, M. (2023). 101 inspiring Winston Churchill quotes that touch on everything—From politics to human nature. Parade. https://parade.com/1034871/marynliles/winston-churchill-quotes/. Accessed 26 July 2024.
[22] Gop, P. (2023). Human conversation training data. Kaggle. https://www.kaggle.com/datasets/projjal1/human-conversation-training-data?resource=download. Accessed 26 July 2024.
[23] Ekman, P. (1992). Are there basic emotions? Psychological Review, 99(3), 550–553. https://doi.org/10.1037/0033-295X.99.3.550
[24] Alon, D., & Ko, J. (2021). GoEmotions: A dataset for fine-grained emotion classification. Google Research Blog. https://research.google/blog/goemotions-a-dataset-for-fine-grained-emotion-classification/. Accessed 26 July 2024.
[25]大穀大. (2023, February 18). ChatGPTが賢くなる！noteの深津さん考案「深津式汎用プロンプト」でChatGPTが劇的に使いやすくなった！. ディレイマニア. https://delaymania.com/202302/webservice/chatgpt-fukatsu-prompt/. Accessed 26 July 2024.
[26] Shao, Y., Li, L., Dai, J., & Qiu, X. (2023). Character-LLM: A trainable agent for role-playing. In Proceedings of the 2023 Conference on Empirical Methods in Natural Language Processing (pp. 13153-13187). Association for Computational Linguistics.
[27] Riedl, M. O., & Young, R. M. (2005). An objective character believability evaluation procedure for multi-agent story generation systems. In: Panayiotopoulos, T. Gratch, J., Aylett, R., Ballin, D., Olivier, P. Rist, T. (eds) Intelligent Virtual Agents. IVA 2005. Lecture Notes in Computer Science, vol 3661. Springer, Berlin, Heidelberg. https://doi.org/10.1007/11550617_24
[28] Liu, B. (2012). Sentiment analysis and opinion mining. Synthesis Lectures on Human Language Technologies, 5(1), 1-167.