臺灣博碩士論文加值系統

近年來由於HIV感染者人數過多，為提升個管師對個案之照護品質，將利用手機軟體使用之特徵值透過迴歸模型方式偵測HIV病患之情緒狀態。根據桃園醫院統計資料指出，平均每年於桃園醫院感染科就診人數約有2400人，而個管師僅有6位，在人力照護上明顯不足，且平均HIV感染者平均每三個月才會回診一次，期間若沒有主動聯繫是很難察覺個案是否有異樣，然而HIV感染者因疾病相關的壓力，容易誘發憂鬱情緒，而憂鬱情緒不易被察覺，因此本研究主要目的為透過手機軟體使用之特徵值偵測HIV病患之情緒狀態。在過去研究中Ballve (2013), Wiese et al. (2015)指出，手機的使用是一種可以追溯的行為特徵，可能與情緒狀態有關。且van Breda et al. (2016)與Shapsough et al. (2016)皆以手機使用特徵值經由機器學習方法達到不錯的預測效果。
因此本研究將透過兩階段分析找出適合之特徵值及預測方法。第一階段分析僅挑選一位受測者之資料並計算使用手機軟體時間之相關特徵值後，透過四種機器學習方法利用均方差評估模型表現，獲得最佳之方法為決策樹模型；第二階段分析將挑選出四位受測者，依照資料收集狀況，分別進行個別受測者資料分析及整體受測者資料分析，並利用均方差及R-sq值判斷模型表現。個別受測者分析中又以手機軟體分群前後，利用決策樹與逐步迴歸分析方法做比較，整體受測者分析中將手機軟體分群後以資料標準化前後做，透過決策樹與逐步迴歸分析方法比較。研究結果發現第二階段分析，個別受測者資料分析中，手機軟體分群前的模型表現較好，兩位受測者之均方差分別為0.0881及0.392；R-sq值分別為99.3%及59.6%。整體受測者資料分析中包含四位受測者之資料，其手機資料標準化前之模型表現較好，均方差為0.4231，R-sq值為40.6%。
從逐步迴歸分析法中獲得模型之重要特徵值，可以了解其每一位受測者心情起伏所影響之手機軟體的使用都不太一樣，因此本研究建議未來可以先對受測者依照心情起伏會影響之手機軟體類型進行分類，可望在未來能夠提升整體受測者之模型表現。

In recent years, there is an increasing number of HIV-infected people. In order to improve the quality of care given by patient managers, the emotional values of HIV patients will be detected through regression models using the characteristic values of mobile phone software. According to statistics from Taoyuan Hospital, the average number of people attending the Infectious Diseases Department in Taoyuan Hospital is about 2,400 annually; however, there are only 6 in the number of patient managers. There is obviously insufficient manpower to provide high quality of care. Furthermore, the average HIV-infected person only visits the doctor once every three months. Whenever there is no active contact during the period, it is difficult to detect when the patient is encountering difficulties and is need of care. Due to disease-related stress, HIV-infected people are prone to induce depression, and depression is not easy to be detected. Therefore, the main purpose of this study is to use the characteristics of mobile phone software. The value detects the emotional state of HIV patients. In past studies, Ballve (2013) and Wiese et al. (2015) pointed out that the use of mobile phones is a traceable behavioral feature that may be related to emotional state. Also, van Breda et al. (2016) with Shapsough et al. (2016) used machine learning methods to achieve good prediction results.
Therefore, this study will find suitable eigenvalues and prediction methods through two-stage analysis. The first stage analysis only selected the data of one subject and calculates the relevant eigenvalues of the time of using the mobile phone software. After using four-machine learning methods, the mean square error evaluation model is used to obtain the best method as the decision tree model. In this stage, analysis is done through selection four subjects and according to the data collection status. Afterwards, analyze the individual subject data and the overall subject data and use the mean square error and R-square value to judge the model performance. In the analysis of individual subjects before and after the mobile phone software grouping, the decision tree was used to compare with the stepwise regression analysis method. In the whole subject analysis, the mobile phone software was grouped and then standardized before and after data. The decision tree and stepwise regression analysis method were also compared. The results of the study found that in the second-stage analysis, the model of the mobile phone software group was better in the analysis of the data of the individual subjects. The mean variance of the two subjects was 0.0881 and 0.392 respectively; the R-SQ values were 99.3% and 59.6 respectively. The analysis of the overall subject data includes information on four subjects. The model before the standardization of mobile phone data performed well, with a mean square error of 0.4231 and an R-SQ of 40.6%.
Obtaining the important eigenvalues of the model from the stepwise regression analysis, we can understand that the use of mobile phone software affected by the mood fluctuation of each subject is not the same. Therefore, this study suggests that the subject’s mood can affect the type of APPs that he uses. The categorization of subjects according to their usage of different mobile phone APP types is expected to improve the overall model performance of the subjects in the future.

摘要 ii
Abstract iii
誌謝 v
第一章 緒論 1
1.1 研究背景與動機 1
1.2 研究目的 2
第二章 文獻探討 4
2.1 人類免疫缺陷病毒(HIV)感染者定義及現行狀況 4
2.2 HIV感染者之憂鬱狀況與診斷 5
2.3 智慧型手機使用與情緒的相關性 7
2.4 模型分析方法 8
2.4.1 決策樹(Decision Tree) 8
2.4.2 隨機森林(Random Forest) 8
2.4.3 支援向量機(Support Vector Machine) 8
2.4.4 K個最近鄰居法(K Nearest Neighbor) 9
2.4.5 逐步迴歸分析法(Stepwise Regression) 9
2.5 文獻小結 9
第三章 研究方法 11
3.1 研究對象與條件限制 11
3.2 研究設備與軟體 11
3.3 實驗流程 11
3.3.1 APP開發 12
3.3.2 實驗收集 16
3.3.3 資料處理 17
3.3.4 模型建構與比較 17
3.3.4.1 第一階段分析 17
3.3.4.2 第二階段分析 19
第四章 研究結果 24
4.1 受測者基本資料 24
4.2 資料收集結果 25
4.3 第一階段資料分析 29
4.3.1 決策樹 29
4.3.2 隨機森林 30
4.3.3 支援向量機 33
4.3.4 K個最近鄰居法 33
4.4 第二階段資料分析 34
4.4.1 個別受測者分析 34
4.4.2 整體受測者分析 42
4.5 研究結果小結 45
4.5.1 第一階段分析 45
4.5.2 第二階段分析 47
第五章 討論 49
5.1 模型適用性 49
5.2 手機軟體分群前後之模型表現 50
5.3 重要特徵值選取對於建模成效影響 50
5.4 探討整體受測者分析之可行性 52
5.5 研究限制 53
第六章 結論 54
6.1 現階段結論 54
6.2 未來規劃 54
參考資料 56
附錄一 受測者同意書內容 62
附錄二 行為調查問卷內容 65

Ballve, M.J.B.I., 2013. How much time do we really spend on our smartphones every day.
Beck, A.T., Steer, R.A., Ball, R. & Ranieri, W.F.J.J.O.P.A., 1996. Comparison of beck depression inventories-ia and-ii in psychiatric outpatients. 67 (3), 588-597.
Betancur, M.N., Lins, L., Oliveira, I.R.D., Brites, C.J.B.J. & Diseases, O.I., 2017. Quality of life, anxiety and depression in patients with hiv/aids who present poor adherence to antiretroviral therapy: A cross-sectional study in salvador, brazil. 21 (5), 507-514.
Breiman, L., Friedman, J.H., Olshen, R.A. & Stone, C.J., 1984. Classification and regression trees.
Breiman, L.J.M.L., 2001. Random forests. 45 (1), 5-32.
Chittaranjan, G., Blom, J. & Gatica-Perez, D., Year. Who's who with big-five: Analyzing and classifying personality traits with smartphonesed.^eds. 2011 15th Annual international symposium on wearable computersIEEE, 29-36.
Control, C.F.D., 2018. Statistics of hiv/aids [online]. https://www.cdc.gov.tw/professional/info.aspx?treeid=7b56e6f932b49b90&nowtreeid=2F13020F8A921CCB&tid=B476404DAD213547 [Accessed Access Date 2019].
Cordell, H.J. & Clayton, D.G.J.T.a.J.O.H.G., 2002. A unified stepwise regression procedure for evaluating the relative effects of polymorphisms within a gene using case/control or family data: Application to hla in type 1 diabetes. 70 (1), 124-141.
Cortes, C. & Vapnik, V.J.M.L., 1995. Support-vector networks. 20 (3), 273-297.
Cover, T.M. & Hart, P.E.J.I.T.O.I.T., 1967. Nearest neighbor pattern classification. 13 (1), 21-27.
Elhai, J.D., Tiamiyu, M.F., Weeks, J.W., Levine, J.C., Picard, K.J., Hall, B.J.J.P. & Differences, I., 2018. Depression and emotion regulation predict objective smartphone use measured over one week. 133, 21-28.
European Aids Clinical Society, E., 2018. Eacs guidelines 2018 version 9.1.
Gonzalez, J.S., Batchelder, A.W., Psaros, C., Safren, S.A. & Syndromes, J.J.O.a.I.D., 2011. Depression and hiv/aids treatment nonadherence: A review and meta-analysis. 58 (2).
Howells, A., Ivtzan, I., Jose Eiroa-Orosa, F. & Studies, J.J.O.H., 2016. Putting the ‘app’in happiness: A randomised controlled trial of a smartphone-based mindfulness intervention to enhance wellbeing. 17 (1), 163-185.
Kroenke, K., Spitzer, R.L. & Williams, J.B.J.J.O.G.I.M., 2001. The phq‐9: Validity of a brief depression severity measure. 16 (9), 606-613.
Krumme, A.A., Kaigamba, F., Binagwaho, A., Murray, M.B., Rich, M.L., Franke, M.F. & Health, J.J.E.C., 2015. Depression, adherence and attrition from care in hiv-infected adults receiving antiretroviral therapy. 69 (3), 284-289.
Kwara, A., Flanigan, T., Carter, E.J.T.I.J.O.T. & Disease, L., 2005. Highly active antiretroviral therapy (haart) in adults with tuberculosis: Current status. 9 (3), 248-257.
Lee, M.-B., Liao, S.-C., Lee, Y.-J., Wu, C.-H., Tseng, M.-C., Gau, S.-F. & Rau, C.-L.J.J.O.T.F.M.A., 2003. Development and verification of validity and reliability of a short screening instrument to identify psychiatric morbidity. 102 (10), 687-694.
Likamwa, R., Liu, Y., Lane, N.D. & Zhong, L., Year. Moodscope: Building a mood sensor from smartphone usage patternsed.^eds. Proceeding of the 11th annual international conference on Mobile systems, applications, and servicesACM, 389-402.
Montgomery, S.A. & #westeur022#sberg, M.J.T.B.J.O.P., 1979. A new depression scale designed to be sensitive to change. 134 (4), 382-389.
Nanni, Maria Giulia Caruso, Rosangela Mitchell, Alex J Meggiolaro, Elena Grassi & Reports, L.J.C.P., 2015. Depression in hiv infected patients: A review. 17 (1), 530.
Newzoo, 2018. Top 50 countries/markets by smartphone users and penetration [online]. https://newzoo.com/insights/rankings/top-50-countries-by-smartphone-penetration-and-users/ [Accessed Access Date
Noryani, M., Sapuan, S.M., Mastura, M.T., Zuhri, M.Y.M., Zainudin, E.S.J.J.O.M.R. & Technology, 2019. Material selection of natural fibre using a stepwise regression model with error analysis.
Saeb, S., Zhang, M., Karr, C.J., Schueller, S.M., Corden, M.E., Kording, K.P. & Mohr, D.C.J.J.O.M.I.R., 2015. Mobile phone sensor correlates of depressive symptom severity in daily-life behavior: An exploratory study. 17 (7).
Shapsough, S., Hesham, A., Elkhorazaty, Y., Zualkernan, I.A. & Aloul, F., Year. Emotion recognition using mobile phonesed.^eds. 2016 IEEE 18th International Conference on e-Health Networking, Applications and Services (Healthcom)IEEE, 1-6.
Silveira, M.P.T., Guttier, M.C., Pinheiro, C.a.T., Pereira, T.V.S., Cruzeiro, A.L.S. & Moreira, L.B.J.R.B.D.P., 2012. Depressive symptoms in hiv-infected patients treated with highly active antiretroviral therapy. 34 (2), 162-167.
Simbayi, L.C., Kalichman, S., Strebel, A., Cloete, A., Henda, N., Mqeketo, A.J.S.S. & Medicine, 2007. Internalized stigma, discrimination, and depression among men and women living with hiv/aids in cape town, south africa. 64 (9), 1823-1831.
Thomas, R., Peethamparan, S.J.C. & Materials, B., 2017. Stepwise regression modeling for compressive strength of alkali-activated concrete. 141, 315-324.
Van Breda, W., Pastor, J., Hoogendoorn, M., Ruwaard, J., Asselbergs, J. & Riper, H., Year. Exploring and comparing machine learning approaches for predicting mood over timeed.^eds. International Conference on Innovation in Medicine and HealthcareSpringer, 37-47.
Who, 2016. Hiv/aids data and statistics [online]. http://www.who.int/hiv/data/en/ [Accessed Access Date 2019].
Wiese, J., Min, J.-K., Hong, J.I. & Zimmerman, J., Year. You never call, you never write: Call and sms logs do not always indicate tie strengthed.^eds. Proceedings of the 18th ACM conference on computer supported cooperative work & social computingACM, 765-774.
Zigmond, A.S. & Snaith, R.P.J.a.P.S., 1983. The hospital anxiety and depression scale. 67 (6), 361-370.