Much structured data of scientific interest can be represented as networks, where sets of nodes or vertices are joined together in pairs by links or edges. Although these networks may belong to different research areas, there is one property that many of them do have in common: the network community structure, which means that there exists densely connected groups of vertices, with only sparser connections between groups. The main goal of community mining is to discover these communities in social networks or other similar information network environments. We face many deficiencies in current community structure discovery methods. First, one similarity metric is typically applied in all networks, without considering the differences in network and application characteristics. Second, many existing methods assume the network information is fully available, and one node only belongs to one cluster. However, in reality, a social network can be huge thus it is hard to access the complete network. It is also common for social entities to belong to multiple communities. Finally, relations between entities are hard to understand in heterogeneous social networks, where multiple types of relations and entities exist. Therefore, the thesis of this research is to tackle these community mining problems, in order to discover and evaluate community structures in social networks from various aspects.

No description available.

Information networks represent relations in data, relationships typically ignored in iid (independent and identically distributed) data. Such networks abound, like coauthorships in bibliometrics, cellphone call graphs in telecommunication, students interactions in Education, etc. A large body of work has been devoted to the analysis of these networks and the discovery of their underlying structure, specifically, finding the communities in them. Communities are groups of nodes in the network that are relatively cohesive within the set compared to the outside. This thesis proposes Top Leaders, a fast and accurate community mining approach for both weighted and unweighted networks. Top Leaders regards a community as a set of followers congregating around a potential leader and works based on a novel measure of closeness inspired by the theory of diffusion of innovations. Moreover, it proposes Meerkat-ED, a specific and practical toolbox for analyzing students interactions in online courses. It applies social network analysis techniques including community mining to evaluate participation of students in asynchronous discussion forums.

No description available.

Public health surveillance is a critical part of understanding, and ultimately influencing, health behaviors. Traditional methods, such as questionnaires and focus groups have significant limitations including cost, delay, and size. Online social media data has the potential to overcome many of the challenges of traditional methods, but its exploitation is not trivial. We develop and apply computational techniques to enable public health surveillance in novel ways and on a larger scale than currently performed.In this regard, we present techniques for mining the who, what, and where of public health surveillance in social media. We show how computational methods can identify health content and conversations in social media, and that people do in fact speak openly about health topics, including those that might be considered private. In addition, we demonstrate how location information can be mined and used to study distributions of various conditions. Finally, and perhaps most importantly, we develop techniques to identify and leverage pertinent social network relationships in public health surveillance. We demonstrate each of these approaches in large data sets of actual social networks spanning blogs, micro-blogs, and video-sharing sites.

No description available.

隨著資訊科技的發展，數位學習的觀念逐漸興起，在二十一世紀強調知識經濟的今天，自主學習及問題解決能力的養成更顯重要，而藉由網路進行問題導向合作學習，學習者可更方便的透過自主學習方式培養問題解決能力。然而學習者在進行網路合作學習的互動期間會接收到大量來自同儕的資訊，有些學習者常會因為無法判斷資訊的正確性，而無法有效選擇、判斷、分析與整合所獲得的資訊，進而觀望同儕或是意見領袖的意見。因此，本研究利用學習者在問題導向網路合作學習歷程中所產生的社會網路互動資料，利用品質Q函數結合基因算法進行社群探勘，並搭配PageRank演算法找尋出每個社群中的較意見領袖，探討採用教師直接進行資訊傳播的一級資訊傳播模式與透過社群意見領袖進行資訊傳播的二級資訊傳播模式對於學習者的學習成效、社會網路互動及團體凝聚力的影響。此外，也探討採用這兩種資訊傳播模式的不同性別及不同人格特質學習者的學習成效、社會網路互動及團體凝聚力是否具有顯著差異。 研究結果發現：(1)在問題導向網路合作學習環境下，採用發佈訊息給意見領袖之二級資訊傳播模式的實驗組學習者，在學習成效上顯著優於教師透過網站公告之一級傳播模式的控制組學習者；(2)在問題導向網路合作學習環境下，採用發佈訊息給意見領袖之二級資訊傳播模式的實驗組女性學習者，在學習成效上顯著優於透過教師網站公告之一級資訊傳播模式的控制組女性學習者，但兩組男性學習者之間則無顯著差異；(3)在問題導向網路合作學習環境下，採用發佈訊息給意見領袖之二級資訊傳播模式的實驗組學習者，在促進同儕互動成效上顯著優於教師透過網站公告之一級傳播模式的控制組學習者;(4)透過品質Q函數結合基因演算法偵測社群，以及使用PageRank找尋社群意見領袖之方法，能精確的協助教師找到問題導向網路合作學習社群之意見領袖。 最後，根據研究結果，本研究提出教學實施及未來研究方向建議，供後續研究參考以進行更深入的探究。 / The concept of e-learning gradually emerges with the development of information technology. In the 21st century when knowledge economy is emphasized, the cultivation of self-directed learning and problem-solving ability becomes more important. Learners with problem-based cooperative learning through networks can more conveniently cultivate the problem-solving ability with self-directed learning. Nonetheless, learners would receive large amount of peer information during the network cooperative learning interaction; some learners therefore could not effectively select, judge, analyze, and integrated the acquired information by judging the accuracy of information to further observe the opinions of peers or opinion leaders. For this reason, learners’ social network interaction data generated in the problem-based network cooperative learning process are proceeded community mining by combining quality function Q and genetic algorithm, and PageRank algorithm is applied to search for the opinion leader in each community in order to discuss the effects of teachers directly proceeding first-order information communication model and second-order information communication model through community opinion leaders on learners’ learning outcome, social network interaction, and group cohesiveness. Furthermore, the effects of such two information communication models on learning outcome, social network interaction, and group cohesiveness of learners with different genders and personality traits are also investigated. The research findings show (1) learners in the experimental group with second-order information communication model by distributing information to opinion leaders, under the problem-based network cooperative learning environment, significantly outperform learners in the control group with first-order communication model through network announcement on the learning outcome; (2) female learners in the experimental group with second-order information communication model by distributing information to opinion leaders, under the problem-based network cooperative learning environment, present remarkably better learning outcome than female learners in the control group with first-order information communication model through network announcement, while no significant difference appears between male learners in both groups; (3) learners in the experimental group with second-order information communication model by distributing information to opinion leaders, under the problem-based network cooperative learning, notably show better peer interaction effectiveness than learners in the control group with first-order communication model through network announcement; and (4) combining quality function Q with genetic algorithm to detect community and applying PageRank to search for community opinion leaders could accurately assist teachers in finding out the problem-based network cooperative learning community opinion leaders. Finally, suggestions for teaching practice and future research, according to the research results, are proposed in this study for successive research.