根據世界衛生組織(WHO)所提供的相關資訊中，在2020年統計中，憂鬱症就佔了全球總人口數的百分之三。而根據今年2023年三月份的統計，憂鬱症人口已經上升到佔全球總人口數的百分之三點八。故本研究提出基於情感、事件與症狀的診斷機器人用以判斷使用者對話中是否具有憂鬱傾向。我們從成大醫院精神科醫師團隊中獲得對病人診斷的錄音檔，並且從中提取出精神科醫師的診斷與對話架構，並且基於此架構來設計本系統的聊天機器人。除此之外我們還從PTT上的憂鬱症版(Prozac)蒐集潛在憂鬱症患者的文章，以利後續的相關訓練與實驗。其中還特別使用到了常識知識庫，一個開源的多語語意知識圖譜工具，讓使用者可以透過知識圖譜的形式，更加清楚理解導致憂鬱的症狀、情感與事件。本系統還結合了簡式量表(BSRS-5)與精神疾病診斷與統計手冊第五版(DSM-5)，經過與精神科醫師的討論後，我們將兩種憂鬱量表問題進行類型整理與精華濃縮轉換成新11題的憂鬱量表，並且將我們使用者的回覆轉換成相對應的分數映射到新的量表中，進而確認使用者的憂鬱傾向。經過實驗過後，我們發現我們所提出的方法在誤差值為1的情況下，準確率可以到達百分之81.8。且相比於傳統問卷填答，聊天機器人能將人類對談之間的溫度，傳遞給使用此系統之人。

According to the information provided by the World Health Organization (WHO), in the 2020 statistics, depression accounted for 3% of the global population. However, according to the statistics from March 2023, the proportion of people with depression has risen to 3.8% of the global population. Therefore, this study proposes a diagnostic chatbot based on emotion, events and symptoms to determine whether the user's conversation exhibits signs of depressive tendencies. We obtained audio recordings of patient diagnoses from the psychiatric department at National Cheng Kung University Hospital, from which we extracted the diagnoses and conversation structure of the psychiatrists. Based on this structure, we designed the chatbot for our system. In addition, we collected articles from the depression board (Prozac) on the PTT forum, targeting potential depressive patients, to facilitate further training and experimentation. We specifically utilized ConceptNet, an open-source multilingual semantic knowledge graph tool, to help users gain a clearer understanding of the symptoms, emotions, and events related to depression through the form of a knowledge graph. Our system also incorporates the Brief Symptom Rating Scale (BSRS-5), as well as the Diagnostic and Statistical Manual of Mental Disorders, Fifth Edition (DSM-5). After discussing with a psychiatrist, we categorized the questions from these two depression scales, condensed and transformed them into a new 11-item depression scale. We then mapped user responses to corresponding scores on the new scale to determine their depressive tendencies.
After conducting experiments, we found that our proposed method can achieve an accuracy rate of 81.8% when the error value is 1. Furthermore, compared to traditional questionnaire surveys, the chatbot is able to convey the warmth of human conversations to users of this system.

[1] World Health Organization, Depressive disorder (depression), WHO, https://www.who.int/news-room/fact-sheets/detail/depression, March. 2023.
[2] 社團法人臺灣憂鬱症防治協會, 關於憂鬱症. https://www.depression.org.tw/knowledge/info.asp?/68.html, April. 2017.
[3] 社團法人臺灣憂鬱症防治協會. (2020, May 13). 台灣憂鬱症就醫現狀與問題. https://www.depression.org.tw/knowledge/info.asp?/68.html
[4] Thong Nguyen, Andrew Yates, Ayah Zirikly, Bart Desmet, and Arman Cohan. Improving the generalizability of depression detection by leveraging clinical questionnaires. In Proceedings of the 60th Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers), pages 8446–8459. 2022.
[5] Anantharaman Karun, S Angel Deborah, S Rajalakshmi, Madhesh Saritha, and RS Milton. SSN_MLRG1@LT-EDI-ACL2022: MultiClass Classification using BERT models for Detecting Depression Signs from Social Media Text. In Proceedings of the Second Workshop on Language Technology for Equality, Diversity and Inclusion. Association for Computational Linguistics. 2022.
[6] Rafał Poswiata and Michał Perełkiewicz. OPI@LT-EDI-ACL2022: Detecting signs of depression from social media text using RoBERTa pretrained language models. In Proceedings of the Second Workshop on Language Technology for Equality, Diversity and Inclusion. 2022.
[7] Lu, IC., Yen Jean, MC., Lei, SM. et al. BSRS-5 (5-item Brief Symptom Rating Scale) scores affect every aspect of quality of life measured by WHOQOL-BREF in healthy workers. Qual Life Res 20, 1469–1475 (2011).
[8] American Psychiatric Association. Diagnostic and statistical manual of mental disorders (5th ed.). https://doi.org/10.1176/appi.books.9780890425596. 2013
[9] Shu-Ling Huang and You-shan Chung. E-HowNet : the Expansion of HowNet. 2008
[10] Sutcliffe P M, BPS monopoles, Int. J. Mod. Phys. A 12 4663–706. 1997.
[11] Tung C, Lu W. Analyzing depression tendency of web posts using an event-driven depression tendency warning model. Artif Intell Med;66:53-62. doi: 10.1016/j.artmed.2015.10.003. Epub 2015 Nov 2. PMID: 26616420. Jan. 2016.
[12] Po-Ming Su, Predicting Web User’s Tendency of Depression by Using Event-Driven Negative Emotion Model, Institute of Medical Informatics National Cheng Kung University 2012
[13] 王壽年，「應用多重消極因素分析模型預測部落客的憂鬱傾向」，國立成功大學資訊工程所碩士論文，2011
[14] He, H., & Choi, J. D. The Stem Cell Hypothesis: Dilemma behind Multi-Task Learning with Transformer Encoders. In Proceedings of the 2021 Conference on Empirical Methods in Natural Language Processing (pp. 5555-5577). Nov. 2021.
[15] de Novais, E.M., Dias Tadeu, T., Paraboni, I. Improved Text Generation Using N-gram Statistics. In: Kuri-Morales, A., Simari, G.R. (eds) Advances in Artificial Intelligence – IBERAMIA 2010. IBERAMIA 2010. Lecture Notes in Computer Science(), vol 6433. Springer, Berlin, Heidelberg. https://doi.org/10.1007/978-3-642-16952-6_32. 2010.
[16] Cheng, H., Shen, Y., Liu, X., He, P., Chen, W., and Gao, J. UnitedQA: A hybrid approach for open domain question answering. In Proceedings of the 59th Annual Meeting of the Association for Computational Linguistics and the 11th In- ternational Joint Conference on Natural Language Processing (Volume 1: Long Papers), pages 3080–3090, Online. Association for Computational Linguistics. 2021.
[17] Chin-Yew Lin. ROUGE: A Package for Automatic Evaluation of Summaries. In Text Summarization Branches Out, pages 74–81, Barcelona, Spain. Association for Computational Linguistics. 2004.
[18] Brown, P. F., Della Pietra, V. J., deSouza, P. V., Lai, J. C., and Mercer,
R. L. Class-based n-gram models of natural language. Computational Linguistics,
18(4):467–480. 1992.
[19] R. Speer, J. Chin and C. Havasi, "ConceptNet 5.5: An open multilingual graph of general knowledge", Proc. AAAI, pp. 4444-4451, 2017.
[20] OpenAI. ChatGPT https://openai.com/chatgpt. 2023.