大型線上社交網絡分析

Analysis of Large-scaled Social Network Services Online

社交網站近年來快速發展，全球最大社交網站Facebook會員數已達十二億人，形成實體人際關係虛擬化的有趣現象。本研究建立分析型模型，以Facebook進行資料蒐集，透過校內及校外進行研究實驗，並分析兩種樣本的結果差異，參與研究實驗合計共226人，其朋友總數為34,681人，並形成627,804條關係連結。研究證明親屬、同學、同事、朋友關係皆會顯著影響群集的連結，且群集類型顯著受到個人背景及個人背景間交互作用效果所影響，唯獨校外樣本上同事關係並不顯著，說明了同事間事實上是存在利害關係的。朋友數量為顯著影響群集效應之因素，但朋友數量卻無顯著影響群集內是否有積極的資訊傳遞角色（例如：意見領袖）。本研究推導出個人與群集建立關係之條件機率、導出關係模型函數，研究也證明隨機交友方式並不容易在社交網絡上形成群集。綜合研究結果將能幫助業界利用社交網站快速準確地切割出目標市場、以及做為行銷、電子商務的重要參考。

Online social service network (SSN) is extremely popular in recent years, such that the Facebook's members have reached 1.2 billion. It creates an interesting phenomenon of virtual-real interpersonal relationships. The sample frame of this research is on the university campus and off the campus. The study also analyses the difference between two samples. Outside the campus are volunteers to participate in the experiment from National Taiwan University Postal Telegraph and Telephone (PTT). The number of participants is 226, which have 34,681 friends and create 627,804 relation links. The study found that all the four kinds of relation types such as family relationships, student relationships, colleague relationships and friend relationships affect the cluster forms ignificantly. Besides, personal relationship has the significant affect on relation types of clusters. The number of friends is the significant factor influencing the cluster effect, but it does not have the significant impact on whether main information senders (i.e. opinion leader) exist in the clusters. Additionally, the study establishes and predicts the probability of connection between the individuals and the clusters. The study also proves that it is not easy to create the clusters on online social network service through making friends randomly. The study develops the relationship function of social service network, and provides true conditional probability of all clusters. The findings will help practitioners precisely create the target market over the online social service network.