世界衛生組織（WHO）預測，未來20年，憂鬱症將成為人類健康的最大威脅。憂鬱症會影響人類的健康和生活作息，例如改變睡眠品質或是飲食習慣等等。根據調查發現，40%的憂鬱症患者會有自殺的想法，其中10%~15％的患者會因為自殺而死。因此，如何早期偵測和預防憂鬱症成為目前主要的研究方向。在這篇論文中，我們分析使用者的生活環境和在社群媒體上的行為，並且提出一個方法去預測使用者的憂鬱程度。我們招募多位頻繁使用臉書的受測者，並且收集他們的臉書資料、問卷調查資料和公開資料，而在這些資料裡，我們特別針對文字在文章中的順序性，在第一階段，利用RNN來計算出每位使用者憂鬱程度的機率值，在第二階段，將機率值加上其他特徵，利用SVM來預測每位使用者的憂鬱程度。實驗結果顯示，從RNN產生的特徵在整個分類器中發揮重要的作用，並且，利用我們的模型來預測使用者的憂鬱程度能達到很好的準確率。

The World Health Organization predicts that depression disorders will be widespread in the next 20 years. These disorders may affect a person’s general health and habits in areas such as altered sleeping and eating patterns in addition to their interpersonal relationships and their work or school life. According to the research, 40% of the patients with depression have suicidal thoughts and 10%~15% die by suicide. Early depression detection and prevention therefore becomes an important issue. In this thesis, a method to predict the depression level of a person by analyzing his/her living environment and behavior in social media will be presented. We recruited several people who use Facebook frequently and collected their Facebook and questionnaire data in addition to some open data to do the analysis. Our method emphasizes the word order of the posts in social media and uses the recurrent neural network (RNN) in the first stage to compute the probabilities of the depression levels. These probabilities are then passed to the second stage with other features to further predict the depression level of the person by support vector machines (SVM). The experiment results show that the feature generated from RNN plays an important role in the prediction and the prediction of the level of depression achieves good accuracy.

[1] World Health Organization. Depression; 2017. (http://www.who.int/mediacentre/factsheets/fs369/en/)
[2] Cavazos-Rehg, P. A., Krauss, M. J., Sowles, S., Connolly, S., Rosas, C., Bharadwaj, M., & Bierut, L. J. (2016). A content analysis of depression-related tweets. Computers in human behavior, 54, 351-357.
[3] Hsieh, Y., & Boland, J. (2010, May). Predicting Processing Difficulty in Chinese Syntactic Ambiguity Resolution: A Parallel Approach. In LSA Annual Meeting Extended Abstracts (Vol. 1, pp. 37-1).
[4] Socher, R., Huval, B., Manning, C. D., & Ng, A. Y. (2012, July). Semantic compositionality through recursive matrix-vector spaces. In Proceedings of the 2012 joint conference on empirical methods in natural language processing and computational natural language learning (pp. 1201-1211). Association for Computational Linguistics.
[5] Shapero, B. G., Black, S. K., Liu, R. T., Klugman, J., Bender, R. E., Abramson, L. Y., & Alloy, L. B. (2014). Stressful life events and depression symptoms: the effect of childhood emotional abuse on stress reactivity. Journal of clinical psychology, 70(3), 209-223.
[6] Zlotnick, C., Kohn, R., Keitner, G., & Della Grotta, S. A. (2000). The relationship between quality of interpersonal relationships and major depressive disorder: findings from the National Comorbidity Survey. Journal of affective disorders, 59(3), 205-215.
[7] American Psychiatric Association. (2013). Diagnostic and statistical manual of mental disorders (DSM-5®). American Psychiatric Pub.
[8] Deng, L., & Yu, D. (2014). Deep learning: methods and applications. Foundations and Trends® in Signal Processing, 7(3–4), 197-387.
[9] Low, L. S. A., Maddage, N. C., Lech, M., Sheeber, L., & Allen, N. (2010, March). Influence of acoustic low-level descriptors in the detection of clinical depression in adolescents. In Acoustics Speech and Signal Processing (ICASSP), 2010 IEEE International Conference on (pp. 5154-5157). IEEE.
[10] Cohn, J. F., Kruez, T. S., Matthews, I., Yang, Y., Nguyen, M. H., Padilla, M. T., ... & De la Torre, F. (2009, September). Detecting depression from facial actions and vocal prosody. In Affective Computing and Intelligent Interaction and Workshops, 2009. ACII 2009. 3rd International Conference on (pp. 1-7). IEEE.
[11] Tekell, J. L., Hoffmann, R., Hendrickse, W., Greene, R. W., Rush, A. J., & Armitage, R. (2005). High frequency EEG activity during sleep: characteristics in schizophrenia and depression. Clinical EEG and neuroscience, 36(1), 25-35.
[12] Andalibi, N., Ozturk, P., & Forte, A. (2015, February). Depression-related imagery on instagram. In Proceedings of the 18th ACM Conference Companion on Computer Supported Cooperative Work & Social Computing (pp. 231-234). ACM.
[13] Tung, C., & Lu, W. (2016). Analyzing depression tendency of web posts using an event-driven depression tendency warning model. Artificial intelligence in medicine, 66, 53-62.
[14] Wang, X., Zhang, C., Ji, Y., Sun, L., Wu, L., & Bao, Z. (2013, April). A depression detection model based on sentiment analysis in micro-blog social network. In Pacific-Asia Conference on Knowledge Discovery and Data Mining(pp. 201-213). Springer, Berlin, Heidelberg.
[15] Park, M., Cha, C., & Cha, M. (2012, August). Depressive moods of users portrayed in Twitter. In Proceedings of the ACM SIGKDD Workshop on healthcare informatics (HI-KDD) (Vol. 2012, pp. 1-8). New York, NY: ACM.
[16] De Choudhury, M., Gamon, M., Counts, S., & Horvitz, E. (2013). Predicting Depression via Social Media. ICWSM, 13, 1-10.
[17] Shen, Y. C., Kuo, T. T., Yeh, I. N., Chen, T. T., & Lin, S. D. (2013, April). Exploiting temporal information in a two-stage classification framework for content-based depression detection. In Pacific-Asia Conference on Knowledge Discovery and Data Mining (pp. 276-288). Springer, Berlin, Heidelberg.
[18] Lai, S., Xu, L., Liu, K., & Zhao, J. (2015, January). Recurrent Convolutional Neural Networks for Text Classification. In AAAI (Vol. 333, pp. 2267-2273).
[19] Arevian, G. (2007, November). Recurrent neural networks for robust real-world text classification. In Proceedings of the IEEE/WIC/ACM International Conference on Web Intelligence (pp. 326-329). IEEE Computer Society.
[20] Hochreiter, S., Bengio, Y., Frasconi, P., & Schmidhuber, J. (2001). Gradient flow in recurrent nets: the difficulty of learning long-term dependencies.
[21] Hochreiter, S., & Schmidhuber, J. (1997). Long short-term memory. Neural computation, 9(8), 1735-1780.
[22] Radloff, L. S. (1977). The CES-D scale: A self-report depression scale for research in the general population. Applied psychological measurement, 1(3), 385-401.
[23] Olsson, G. I., & Knorring, A. L. (1999). Adolescent depression: prevalence in Swedish high‐school students. Acta Psychiatrica Scandinavica, 99(5), 324-331.
[24] Ramirez-Esparza, N., Chung, C. K., Kacewicz, E., & Pennebaker, J. W. (2008, March). The Psychology of Word Use in Depression Forums in English and in Spanish: Texting Two Text Analytic Approaches. In ICWSM.
[25] Ku, L. W., & Chen, H. H. (2007). Mining opinions from the Web: Beyond relevance retrieval. Journal of the Association for Information Science and Technology, 58(12), 1838-1850.
[26] Cheng, C. M., Chen, H. C., Chan, Y. C., Su, Y. C., & Tseng, C. C. (2013). Taiwan corpora of Chinese emotions and relevant psychophysiological data—Normative Data for Chinese Jokes. Chin. J. Psychol, 55, 555-569.
[27] Bengio, Y., Ducharme, R., Vincent, P., & Jauvin, C. (2003). A neural probabilistic language model. Journal of machine learning research, 3(Feb), 1137-1155.
[28] Guyon, I., Weston, J., Barnhill, S., & Vapnik, V. (2002). Gene selection for cancer classification using support vector machines. Machine learning, 46(1), 389-422.
[29] Yang, H. J. (2002). A follow-up study of depressive disorders and depressive symptoms in adolescents. National Taiwan University, Taipei, Republic of China.
[30] Lodhi, H., Saunders, C., Shawe-Taylor, J., Cristianini, N., & Watkins, C. (2002). Text classification using string kernels. Journal of Machine Learning Research, 2(Feb), 419-444.
[31] Sun, J. (2012). ‘Jieba’Chinese word segmentation tool.
[32] Mikolov, T., Chen, K., Corrado, G., & Dean, J. (2013). Efficient estimation of word representations in vector space. arXiv preprint arXiv:1301.3781.