動態社群網路中的演化樣式探勘

Mining Evolution Pattens in Dynamic Social Networks

華晟彣(Sheng-Wen Hua)

指導教授 ： 陳以錚

淡江大學/工學院/資訊工程學系碩士在職專班/碩士(2016年)

https://doi.org/10.6846/TKU.2016.00763

摘要

近年來，由於社群網站和應用程式的普及，大部分的研究都集中在結構化的社群網路分析中。現在許多的社會現象，在實際應用中，可以由社群網路中提取有意義的社交模式。然而，建立或刪除過時的用戶和關係，這通常隨著時間而演變，這樣的動態特性絕對會增加模式探勘的複雜度，過去一些研究已經提出在動態網絡的探索模式，這些研究通常將動態社群網絡處理成為一系列的關係圖，並從每個關係圖中探勘出有意義的子圖形（聚集），最後整合探勘結果成為一個序列，事實上，這些序列子圖或聚集可能不足以揭示個人的演化特徵; 以使用者的角度而言，隨著時間變化它仍然是難以觀察出與其他個體相互作用的演化。在本論文中，我們採用一個新的的表示法來表達動態社群網絡和一個新型的模式，以及演化特徵，並在一個動態社群網絡中獲取互動演化樣式。 此外，提出一個新的演算法Evolution Characteristic Miner (ECM),有效的提昇探勘效率與維護演化特徵。ECM還採用了一些優化策略，有效地減少搜尋空間以提高性能。使用虛擬與實際資料庫的實驗結果表示ECM於動態社群網路中探勘互動演化樣式的效率，並擁有優雅的可擴展性。最後，我們應用在真實數據集ECM顯示演變特徵挖掘的實用性。

關鍵字

樣式探勘 ； 動態社群網路 ； 社群網路分析

並列摘要

Recently, due to the popularity of social website and APP, considerable attention has been paid to the analysis of structure in social network. Many potential social phenomena, in practice, can be extracted by discovering significant pattern in social network. Clearly, the network usually evolves with time; some new users and relationships are established, and some obsolete ones are removed. This dynamic feature definitely increases the complexity of pattern discovery. Several prior studies have proposed to discover pattern in dynamic network. These works generally treat the dynamic social network as a series of graphs and discover the significant subgraphs (clusters) from each graph, and then integrate the discovered results into a series. However, in fact, the series of subgraphs or clusters may be insufficient to reveal the evolving characteristics of individuals; in user’s point of view, it is still difficult to observe the evolution of interaction with other individuals over time. In this dissertation, we introduce a new representation to express the dynamic social network and a new type of pattern, evolution characteristic, to capture the interaction evolutions in a dynamic social network. Furthermore, a novel algorithm, Evolution Characteristic Miner (ECM), is developed to discover and maintain the evolution characteristics efficiently. ECM also employs some pruning strategies to effectively reduce the search space to improve the performance. The experimental results on synthetic and real datasets show the efficiency of ECM for extracting interaction evolution in dynamic networks, and possess graceful scalability. Finally, we apply ECM on real datasets to show the practicability of evolution characteristic mining.

並列關鍵字

Pattern Mining ； Dynamic Social Network ； Social Network Analysis

參考文獻

[1] Allen, J. 1983. Maintaining Knowledge about Temporal Intervals. Communications of ACM 26(11) 832-843.

[9] Hopper, F. 2002. Finding informative rules in interval sequences. Intelligent Data Analysis 6(3). 237-255.

[10] Inokuchi, A., and Washio, T. 2008. A fast method to mine frequent subsequences from graph sequence data. In Proceeding of the 8th IEEE International Conference on Data Mining (ICDM’08). 303-312.

[12] Kim, M., and Han, J. 2009. A Particle-and-Density based Evolutionary Clustering Method for Dynamic Networks. In Proceedings of the 35th International Conference on Very Large Data Bases (VLDB’09).

[13] Lahiri, M., and Berger-Wolf, T. 2007. Structure Prediction in Temporal Networks using Frequent Subgraphs. In Proceedings of the 2007 IEEE Symposium on Computational Intelligence and Data Mining (CIDM’07). 35-42.