網路社群的蓬勃發展有目共睹，但管理者需能掌握社群動態，才能維護社群的長遠發展。然而，前人研究多著力於社群成員影響力之預測與訊息傳遞方式，對於個別成員或總體動態發展變化，則少有著墨。有鑑於此，本研究針對網路社群成員的動態發展，提出一套有效的預測方法。首先，我們利用節點分支度、中心性等七種指標，作為描述社群成員特質的依據。接著，利用k-means演算法對社群成員進行分群，根據群心來了解成員的典型特質，並據以建立成員狀態，本研究提出5大類型成員：外向型、受關注型、積極互動型、一般型、邊緣型。接著，利用馬可夫鏈模型產生預測模型，透過建立狀態轉換矩陣，了解社群成員未來可能的狀態轉變。其次，我們也透過循序樣式探勘概念，建立社群的常見狀態變化循序樣式，以預測成員未來的狀態變遷。此外，本研究亦以滑動視窗概念建立迴歸模型，期能預測討論區之發文量。
    為驗證提出方法之有效性，本研究蒐集PTT不同討論區的實際資料來進行實驗。結果顯示，以馬可夫鏈為基礎之預測方法準確率可為68.5%～74%，且隨著order值增加而上升；以循序樣式為基礎之方法準確率，則會隨著實驗週數增加而增加。此外，透過線性迴歸對社群發文量進行預測時，結果顯示與實際值具有良好的相關度。綜合上述結果，驗證本研究提出方法確實有助於網路社群成員之動態發展預測，能作為管理當局規劃未來的發展之重要參考依據。

With the advancement of social network, social network manager must care development of social network, then we can maintain long development of social network. However, previous studies have focused on prediction of the influence of members and method of message delivery. Have not focused on change of dynamic development of members or social network. Because of this, this research found an effective method to predict members of dynamic development. First, we used 7 kinds of attributes with Node degree、Centrality etc. as a basis for describing the characteristics of members. Next, we used k-means clustering algorithm to separate social network members. According to the cluster centroids, we can understand typical characteristics of members then create members status. This research divided into five types of members : Extrovert、Be-Concerned、Positive Interaction、Borderline、Neutrals. Then, we use the Markov Chain Model to generate predictive model, understanding the possible changes of members in all future through the establishment of state transition matrix. Second, build Common state changes of social network sequential pattern based on the concepts of sequential pattern. To predict changing of the future status of members. In addition, this research establish a regression model based on the concept of sliding windows. Hope can predict numbers of published article of forums.
To verify the effectiveness of the proposed method, This research data download from PTT forum are used for experiment, and the results show that based on the Markov chain, the accuracy rate can be 68% to 74%, and be increased as the value of order increases; Based on the Sequential Pattern, the accuracy rate can be increased with increased number of weeks of experiment. In addition, predict numbers of publish article through linear regression, the results has good correlated with value of actual. 
The above results that demonstrate the effectiveness of the proposed method can understand dynamic development of social network member’s prediction, it can help social network manager planning the future development and policy.

目錄 
=================================== 
第一章　緒論                                           1
第二章　相關技術與背景知識                              4
  2.1	網路社群                                       4
  2.2	馬可夫鏈	                                      5
  2.3	社群網路分析指標	                              6
  2.4	網路社群成員分類	                              8
第三章　網路社群成員動態發展預測方法	             10
  3.1	社群成員網路社群指標	                     10
  3.2	社群成員角色類型	                             12
  3.3	建立模型	                                     15
    3.3.1 以馬可夫鏈為基礎之模型	                     15
    3.3.2 使用循序樣式進行社群成員影響力狀態變遷與預測   19
    3.3.3 運用線性迴歸進行社群發文量之預測	             20
第四章　實驗結果	                                     22
  4.1	實驗設定	                                     22
  4.2	實驗結果	                                     23
    4.2.1 運用馬可夫轉換矩陣進行預測	             23
    4.2.2 社群成員類型長期發展預測	             25
    4.2.3 運用循序樣式概念進行使用者狀態預測	     30
    4.2.4 運用線性迴歸進行社群發文量之預測	             35
第五章　結論與未來工作	                             39
參考文獻	                                             41
附錄	                                             45

表目錄 
=============================== 
表3-1 : 本研究使用之社群成員特性描述指標	11
表3-2 : 根據2016/01/01～2016/08/31資料所建立之各群群心屬性值	14
表3-3：本研究整理	15
表3-4: 說明如何建立馬可夫鏈轉換矩陣之範例	16
表3-5 : 第1列是週數，第2,3列是分群後社群成員在不同週數所屬的群集	17
表3-6 : 每位使用者出現的所有pattern	17
表3-7: 將表3-6的結果建立成馬可夫變遷矩陣	17
表3-8：政黑版8個月之資料所建構之order-1變遷矩陣	18
表3-9：Stock版8個月之資料所建構之order-1變遷矩陣	18
表3-10：IPhone版8個月之資料所建構之order-1變遷矩陣	18
表3-11: 影響力狀態變遷序列分析	20
表4-1 : 政黑版-利用前26週的資料建立之馬可夫轉移矩陣	27
表4-2 : Stock版-利用前30週的資料建立之馬可夫轉移矩陣	27
表4-3 : IPhone版-利用前26週的資料建立之馬可夫轉移矩陣	27
表4-4 : 政黑版-在長期發展下，不同特質之分佈比例預測	28
表4-5 : Stock版-在長期發展下，不同特質之分佈比例預測	28
表4-6 : IPhone版-在長期發展下，不同特質之分佈比例預測	28
表4-7：政黑版4w:17w中資料比對正確的結果樣式	32
表4-8：政黑版16w:5w中資料比對正確的結果樣式	32
表4-9：政黑版-將圖4-10整理成各群群心距離對照表	34
表4-10：Stock版-將圖4-11整理成各群群心距離對照表	34
表4-11：IPhone版-將圖4-12整理成各群群心距離對照表	34

圖目錄 
==================================== 
圖2-1: 根據γ和ρ係數的散點圖結果，將社群使用者分成不同特質之角色(Shafiq, et al. 2013)	9
圖3-1: gap統計合適社群成員分群之群數	12
圖3-2: 社群成員影響力狀態變遷圖範例	19
圖4-1 : 政黑版-使用不同週數和lookorder值搭配實驗結果	24
圖4-2 : Stock版-使用不同週數和lookorder值搭配實驗結果	25
圖4-3 : IPhone版-使用不同週數和lookorder值搭配實驗結果	25
圖4-4 : 政黑版-角色類型預測分布情況與實際週數資料比對	28
圖4-5 : Stock版-角色類型預測分布情況與實際週數資料比對	29
圖4-6 : IPhone版-角色類型預測分布情況與實際週數資料比對	29
圖4-7 ：政黑版-不同週數時間切點之實驗結果	31
圖4-8 ：Stock版-不同週數時間切點之實驗結果	31
圖4-9 ：IPhone版-不同週數時間切點之實驗結果	31
圖4-10 ：政黑版-各群群心彼此間群心距離比較圖	33
圖4-11 ：Stock版-各群群心彼此間群心距離比較圖	33
圖4-12 ：IPhone版-各群群心彼此間群心距離比較圖	34
圖4-13 ：使用不同週數所建立之預測結果與實際資料間之相關係數	36
圖4-14 ：預測波型與實際資料波型之比對錯誤次數	36
圖4-15 ：政黑版-發文量預測結果	37
圖4-16：Stock版-發文量預測結果	37
圖4-17 ：IPhone版-發文量預測結果	37
圖4-18 ：「網絡規模」、「網絡密度」、「群聚係數」使用不同週數所建立之預測結果與實際資料間之相關係數	38
圖4-19 ：「網絡規模」、「網絡密度」、「群聚係數」、「平均最短路徑長」使用不同週數所建立之預測結果與實際資料間之相關係數	38

[1]Binder, Jens, Andrew Howes, and Alistair Sutcliffe. "The problem of conflicting social spheres: effects of network structure on experienced tension in social network sites." Proceedings of the SIGCHI conference on human factors in computing systems. ACM, 2009.
[2]Benevenuto, Fabrício, et al. "Characterizing user behavior in online social networks." Proceedings of the 9th ACM SIGCOMM conference on Internet measurement conference. ACM, 2009.
[3]Bodendorf, Freimut, and Carolin Kaiser. "Detecting opinion leaders and trends in online communities." Digital Society, 2010. ICDS'10. Fourth International Conference on. IEEE, 2010.
[4]Cheng, C-J., et al. "Customer lifetime value prediction by a Markov chain based data mining model: Application to an auto repair and maintenance company in Taiwan." Scientia Iranica 19.3 (2012): 849-855.
[5]Chan, S. L., W. H. Ip, and Vincent Cho. "A model for predicting customer value from perspectives of product attractiveness and marketing strategy." Expert Systems with Applications 37.2 (2010): 1207-1215.
[6]Ellison, Nicole B. "Social network sites: Definition, history, and scholarship." Journal of Computer‐Mediated Communication 13.1 (2007): 210-230.
[7]Ellison, Nicole B., Charles Steinfield, and Cliff Lampe. "The benefits of Facebook “friends:” Social capital and college students’ use of online social network sites." Journal of Computer‐Mediated Communication 12.4 (2007): 1143-1168.
[8]Franks, H., Griffiths, N. and Anand, S. S., "Learning agent influence in MAS with complex social networks," Auton Agent Multi-Agent Syst (2014) 28:836-866. 
[9]Genter, K. and P. Stone. Ad hoc teamwork behaviors for influencing a flock. Acta Polytechnica, 56(1):18-26, 2016.
[10]Girvan, Michelle, and Mark EJ Newman. "Community structure in social and biological networks." Proceedings of the national academy of sciences 99.12 (2002): 7821-7826.
[11]Ha, Sung Ho, Sung Min Bae, and Sang Chan Park. "Customer's time-variant purchase behavior and corresponding marketing strategies: an online retailer's case." Computers & Industrial Engineering 43.4 (2002): 801-820.
[12]Hoppe, Bruce, and Claire Reinelt. "Social network analysis and the evaluation of leadership networks." The Leadership Quarterly 21.4 (2010): 600-619.
[13]Huang, Baocheng, Guang Yu, and Hamid Reza Karimi. "The finding and dynamic detection of opinion leaders in social network." Mathematical Problems in Engineering 2014 (2014).
[14]Kempe, David, Jon Kleinberg, and Éva Tardos. "Maximizing the spread of influence through a social network." Proceedings of the ninth ACM SIGKDD international conference on Knowledge discovery and data mining. ACM, 2003.
[15]Liu., Y., et al., "Modeling What Friendship Patterns on Facebook Reveal About Personality and Social Capital," ACM Transactions on Computer-Human Interaction, Vol. 21, No. 3, Article 17, May 2014. pp. 16-36.
[16]Mehra, Ajay, et al. "The social network ties of group leaders: Implications for group performance and leader reputation." Organization science 17.1 (2006): 64-79.
[17]Kaiser, Xiaodan, et al. "Identifying opinion leaders in the blogosphere." Proceedings of the sixteenth ACM conference on Conference on information and knowledge management. ACM, 2007.
[18]Shafiq, M. Zubair, et al. "Identifying leaders and followers in online social networks." IEEE Journal on Selected Areas in Communications 31.9 (2013): 618-628.
[19]Subbian, K., et al., "Mining Influencers Using Information Flows in Social Streams," ACM Transactions on Knowledge Discovery from Data, Vol. 10, No. 3, Article 26, January 2016.
[20]Tang., X. and Yang, C. C., "Ranking User Influence in Healthcare Social Media," ACM Transactions on Intelligent Systems and Technology, Vol. 3, No. 4, Article 73, September 2012. pp. 73-93.
[21]Tavakolifard, Mozhgan, and Kevin C. Almeroth. "Social computing: an intersection of recommender systems, trust/reputation systems, and social networks." Network, IEEE 26.4 (2012): 53-58.
[22]Tibshirani, Robert, Guenther Walther, and Trevor Hastie. "Estimating the number of clusters in a data set via the gap statistic." Journal of the Royal Statistical Society: Series B (Statistical Methodology) 63.2 (2001): 411-423.
[23]Wilson, Christo, et al. "User interactions in social networks and their implications." Proceedings of the 4th ACM European conference on Computer systems. Acm, 2009.
[24]Wittie, Mike P., et al. "Exploiting locality of interest in online social networks." Proceedings of the 6th International COnference. ACM, 2010.
[25]Wondracek, Gilbert, et al. "A practical attack to de-anonymize social network users." Security and Privacy (SP), 2010 IEEE Symposium on. IEEE, 2010.
[26]Zeng, Fue, Li Huang, and Wenyu Dou. "Social factors in user perceptions and responses to advertising in online social networking communities." Journal of interactive advertising 10.1 (2009): 1-13.
[27]Zhang, H., et al., "Misinformation in Online Social Networks: Detect Them All with a Limited Budget," ACM Transactions on Information Systems, Vol. 34, No. 3, Article 18, April 2016. pp. 18-42.
[28]林冠呈. 運用分群演算法找出Facebook粉絲團的意見領袖. 朝陽科技大學資訊工程學系碩士論文, 2015.
[29]林倉億. 『轉移矩陣』二三事(3)：馬可夫鏈穩定狀態的判別. HPM通訊第十七卷第七八期, 2014.
[30]Mengqi(2016)線性代數拾遺：特徵值與特徵向量。2017年2月21日
取自 http://mengqi92.github.io/2016/07/01/linear-algebra-6/#more