公務人員關注議題之文字探勘：以PTT公職板為例

A Computational Text Analysis on the Core Issues for Public Servants: Evidence from the PTT

10.7014/SRMA.2020100005

王貿(Mao Wang)

公務人員 ； 文字探勘 ； 批踢踢實業坊 ； 內部顧客 ； 結構式主題模型 ； public servant ； text mining ； PTT ； internal customer ； structural topic model

調查研究-方法與應用

45期（2020 / 10 / 01）

119 - 154

繁體中文

以往政府並不重視基層公務人員的意見，除了缺乏溝通平臺外，更常見的是許多意見交流只存在於個人層次的人際管道。然而，隨著各種網路論壇與社群的興起，亦產生了以公務人員為主要使用者的討論社群，內容則是與公職生涯有關的各種議題。從內部顧客管理的觀點，理解公務人員的疑問與需求，正是公務人力資源管理的關鍵。基此，本研究以「批踢踢實業坊」（PTT）中的「公職板」為研究標的，蒐集該板內屬於提問性質之發文共20,616篇，利用文字探勘之方法，描繪公務人員關注之議題面向。依結構式主題模型分析之結果發現，發文可以歸納為13個主題：福利、活動、互動、請假、備考、採購、就任、錄取、改革、調任、志願、考績、詢缺。繼而深入探討主題內容並對照實務運作，另以主題網絡呈現主題間之關聯。最後，依研究結果提出相關建議，供政府人事主管機關與未來研究參考。

Background: The government used to ignore the voice of public servants from the bottom levels, the reasons being not only a lack of communication channels but also the fact that such information often exists exclusively within personal networks. However, along with the rise of online forums and social media, there are some communities, discussing all kinds of issues related to public jobs, mainly for public servants. One key element of public personnel management is to realize the problems raised by and needs of the internal customer, namely civil service, and thus this study is to investigate what issues are most discussed online by public servants. Methods and Data: Survey is a common way to understand the voice of internal customers, but it is also accompanied by certain problems, such as a narrower survey framework decided by the survey administrator, and respondents' fear of identity disclosure to their supervisors. Using textual data from online forum posts, without the downside of the survey method, and exploiting the techniques of computational text analysis are helpful to depict the issues public servants truly care about. This article focuses on textual data from the PublicServan Board (公職板)of the PTT (like a Taiwanese version of Reddit), collects 20,616 posts which ask a question, and summarizes the content of posts through an unsupervised machine learning method. Results: Based on the result of a structural topic model, this article examines these posts and can be clustered into 13 topics: benefits, activities, interactions, leave, examinations, procurement, reports, recruitment, reform, transfers, job rankings, appraisals, and position seeking, and compares them to the government practices we recognize, and visualizes the relationships between the topics. In this corpus, the amount of posts on each topic varies. The top four topics include: job ranking (11.15%), position seeking (10.32%), recruitment (9.61%), and interactions (9.43%), and the results reveal that qualified people tend to raise questions about how to find an ＂appropriate＂ agency and a ＂suitable＂ position. This finding is in line with the fact that the numbers of posts are higher every March and September because that is when the results of the public service exams are announced. Apart from discerning different topics, there are some discrepancies and contradictions between the findings from these textual data and the annual survey administered by the Directorate-General of Personnel Administration, Executive Yuan (行政院人事行政總處). The high turnover rate is attributed to dissatisfaction with promotion by the authority, but the relationship between turnover and dissatisfaction with promotion rarely shows evidence from the posts we collect. Interactions with line managers and colleagues in the workplace often result in doubts and complaints which we can discover from textual data, yet no numeric data explicitly indicate this situation in the annual survey. Finally, the topic network adds more contextual information to specify topic categories and uncover their relationships. Conclusions: Computational text analysis does not replace traditional ways of comprehending public servants, but can complement and confirm what we know about public servants. This type of analysis provides not merely higher or lower numbers but a meaningful context, and can, more importantly, help enhance interpretations. Some suggestions are made for the reference of the government personnel authority and future research agendas.

1. 汪明生, Ming-Shen,陳正料, Cheng-Liaou,林錦郎, Chin-Lang(2005)。政府內部顧客參與行為之研究：以都市垃圾清運業務委外案為例。公共行政學報，15，81-129。
   連結：
2. 曹開明, Kai-Ming,劉大華, Ta-Hau(2019)。網路論述的語藝生態學初探：以網路社群討論軍中同志議題為例。傳播研究與實踐，9(1)，65-103。
   連結：
3. 陳世榮, Roger(2015)。社會科學研究中的文字探勘應用：以文意為基礎的文件分類及其問題。人文及社會科學集刊，27(4)，683-718。
   連結：
4. 傅文成, Wen-Cheng,陶聖屏, Sheng-Ping(2018)。以大數據觀點探索網路謠言的「網路模因」傳播模式。中華傳播學刊，33，99-135。
   連結：
5. 管瓊瑛, Chiung-Ying,謝寧, Ning,陳潔, Chieh,張桂萍, Kuei-Ping,高翊瑋, Yi-Wei,謝邦昌, Ben-Chang,張嘉芳, Chia-Fang,張耀懋, Yao-Mao(2017)。長期照顧政策是照顧老人還是失能者？─以蔡英文臉書為例探勘民眾認知。臺灣公共衛生雜誌，36(55)，511-520。
   連結：
6. 謝吉隆, Ji-Lung,楊苾淳, Bi-Chun(2018)。從「應變自然」到「社會應變」：以文字探勘方法檢視國內風災新聞的報導演變。教育資料與圖書館學，55(3)，285-318。
   連結：
7. 蘇偉業, Bennis Wai-Yip(2018)。我國政府內部人力市場行為之初探：個人利益與組織利益之權衡。文官制度季刊，10(1)，21-58。
   連結：
8. Alvarez, R. Michael(ed.)(2016).Computational Social Science: Discovery and Prediction.New York:Cambridge University Press.
9. Blei, David M.,Ng, Andrew Y.,Jordan, Michael I.(2003).Latent Dirichlet Allocation.Journal of Machine Learning Research,3,993-1022.
10. Denhardt, Janet V.,Denhardt, Robert B.(2011).The New Public Service: Serving, Not Steering.Armonk, NY:M.E. Sharpe.
11. Denny, Matthew J.,Spirling, Arthur(2018).Text Preprocessing For Unsupervised Learning: Why It Matters, When It Misleads, And What To Do About It.Political Analysis,26(2),168-189.
12. DiMaggio, Paul,Nag, Manish,Blei, David(2013).Exploiting Affinities between Topic Modeling and the Sociological Perspective on Culture: Application to Newspaper Coverage of U.S. Government Arts Funding.Poetics,41(6),570-606.
13. Grimmer, Justin,Stewart, Brandon M.(2017).Text as Data: The Promise and Pitfalls of Automatic Content Analysis Methods for Political Texts.Political Analysis,21(3),267-297.
14. Hirschman, Albert O.(1970).Exit, Voice, and Loyalty: Responses to Decline in Firms, Organizations, and States.Cambridge, MA:Harvard University Press.
15. Hollibaugh, Gary E.(2019).The Use of Text as Data Methods in Public Administration: A Review and an Application to Agency Priorities.Journal of Public Administration Research and Theory,29(3),474-490.
16. Isoaho, Karoliina,Gritsenko, Daria,Mäkelä, Eetu(2019).Topic Modeling and Text Analysis for Qualitative Policy Research.Policy Studies Journal
17. Laver, Michael,Benoit, Kenneth,Garry, John(2003).Extracting Policy Positions from Political Texts Using Words as Data.American Political Science Review,97(2),311-331.
18. OPM, 2016, “Federal Employee Viewpoint Survey Technical Report.” https://www.opm.gov/fevs/reports/technical-reports/technical-report/technical-report/2016/2016-technical-report.pdf (Date visited: September 15, 2019).
19. Osborne, David,Gaebler, Ted(1992).Reinventing Government: How the Entrepreneurial Spirit Is Transforming the Public Sector.Reading, MA:Addison-Wesley.
20. Roberts, Margaret E.,Stewart, Brandon M.,Tingley, Dustin(2019).stm: R Package for Structural Topic Models.Journal of Statistical Software,91(2),1-40.
21. Roberts, Margaret E.,Stewart, Brandon M.,Tingley, Dustin,Lucas, Christopher,LederLuis, Jetson,Gadarian, Shana Kushner,lbertson, Bethany ,Rand, David G.(2014).Struc-tural Topic Models for Open-Ended Survey Responses.American Journal of Political Science,58(4),1064-1082.
22. Tang, Jian,Meng, Zhaoshi,Nguyen, XuanLong,Mei, Qiaozhu,Zhang, Ming(2014).Understanding the Limiting Factors of Topic Modeling via Posterior Contraction Analysis.The 31st International Conference on Machine Learning,Beijing, China:
23. Tepper, Bennett J.(2000).Consequences of Abusive Supervision.Academy of Management Journal,43(2),178-190.
24. Tepper, Bennett J.(2007).Abusive Supervision in Work Organizations: Review, Synthesis, and Research Agenda.Journal of Management,33(3),261-289.
25. Walker, Richard M.,Chandra, Yanto,Zhang, Jiasheng,Arjen van Witteloostuijn(2019).Topic Modeling the Research Practice Gap in Public Administration.Public Administration Review,79(6),931-937.
26. Welbers, Kasper,Wouter Van Atteveldt,Benoit, Kenneth(2017).Text Analysis in R.Communication Methods and Measures,11(4),245-265.
27. Zuo, Zhiya,Qian, Haifeng,Zhao, Kang(2019).Understanding the Field of Public Affairs through the Lens of Ranked Ph.D. Programs in the United States.Policy Studies Journal,47(S1),S159-S180.
28. 考試院=Examination Yuan of ROC(2017)。考試院，2017，考試院第 12 屆第 157 次會議紀錄（https://ws.exam.gov.tw/Download.ashx?u。https://ws.exam.gov.tw/Download.ashx?u=LzAwMS9VcGxvYWQvMS9yZWxmaWxlLzg5NTcvMzA4NTgvZTgzOTk1ZjctMDUyZi00ODdjLWExMjYtMzU2YzIyZGU3NjdkLnBkZg%3d%3d&n=NzEwMTI5NDQ4NzEucGRm&icon=.pdf
29. 李天申, Tian-Sheng(2013)。我國推動公務人員考績丙等比例的政策衝突分析。文官制度季刊，5(1)，21-47。
30. 陳敦源, Don-yun,黃東益, Tongyi,蕭乃沂, Naiyi,郭思禹, Szu-yu(2006)。官僚回應性與內部顧客關係管理：台北市政府市長信箱個案研究。行政暨政策學報，42，143-182。
31. 黃東益, Tong-Yi(2009)。電子化政府的影響評估：內部顧客的觀點。文官制度季刊，1(3)，25-53。

1. 蔡金田,楊金君,沈秋宏(2023)。應用R語言文字探勘技術進行教育研究之探究。學校行政，143，120-150。
2. 曾繁絹,白佩宜(2023)。以語料庫分析社群媒體探討地方觀光吸引力－臺南市政府臉書專頁為例。圖書資訊學刊，21(2)，89-118。
3. 陳明吉,高嘉璘,洪志興,朱芳妮(2023)。社群媒體情緒與房市交易資訊關係之檢視。應用經濟論叢，113，85-138。
4. 林巧敏,李育賢(2023)。臺灣網路論壇關注之檔案事件主題及其情感分析。圖書資訊學研究，17(2)，71-107。
5. 沈秋宏,沈芷嫣(2022)。國民小學實施課程標準時期教育領導研究趨勢之探究－應用文字探勘技術分析。教育理論與實踐學刊，45，1-33。
6. 釋淨如,蔡金田,沈秋宏(2022)。國民小學實施九年一貫課程期間教育領導趨勢之探究－應用文字探勘技術分析。學校行政，138，74-104。
7. (2023)。大眾與社群媒體如何形塑公眾對再生能源的認知—以框架理論檢視新聞報導及PTT論壇為例。臺灣能源期刊，10(3)，213-234。