中華傳播學刊的引用網絡分析：指數隨機圖模型與動態模型之取徑

Abstract

網絡資料中存在的關係依賴性，是社會網絡資料的特殊之處，也是分析時須處理的核心議題。有鑒於此，網絡研究需要更完整的模型以研究變項間的關係。近年來，隨著統計方法與電腦模擬技術的進展，關於指數隨機圖模型（p*）、動態模型的研究受到高度的關注。指數隨機圖模型基於特定的依賴性假定，能妥善處理網絡結構與行動者屬性間的關係，並且有效估計特定網絡結構出現的機率。而動態模型則易於區分行動者屬性和網絡結構間的因果關係。 本文透過社會網絡的觀點，分析中華傳播學刊的引用網絡。採用指數隨機圖模型來分析靜態網絡，並運用動態模型來分析動態網絡，期望全面了解學刊的引用情況。本文運用內容分析法，針對中華傳播學刊21期的中文引用文獻進行編碼，並使用UCINET、PNet、StOCNET等軟體進行檢視。 結果發現：一、在文章網絡中較存在偏好連結（preferential attachment）的現象。二、在作者網絡中較存在同質性的現象。三、社會影響與社會選擇效果，有部分的顯著效果。四、初階的指數隨機圖模型之結果，可能會與高階模型的結果不同，而越初階的模型越有可能產生退化的問題。

The dependence between dyads is one of the distinctive features of social network data, and it is also a key issue of this field. To deal with this problem, the researchers need more complete models. In recent years, with the progress of the statistical methods and the simulate technique of computers, there are more researches lay stress on exponential random graph model (p*) and dynamic model. On one hand, the dependence assumption is the basic concern of exponential random graph model. This model can cope with the relationships between the network configuration and actor’s attributes, and effectively estimate the probability of network configuration. On the other hand, dynamic model can cope with the causal relationship between the network configuration and actor’s attributes. In this paper, we use the perspective of social network analysis to analyze the citation network of Chinese Journal of Communication Research. We introduce exponential random graph model and dynamic model to analyze the static networks and dynamic networks respectively, and hoping to have a comprehensive understanding of the citation networks. The paper uses content analysis to analyze the citation networks, encoding twenty one volumes of Chinese Journal of Communication Research, and using the software as UCINET, PNet, and StOCNET for viewing. Above-mentioned procedures could demonstrate some outcomes as following. First of all, the phenomenon of preferential attachment happened more frequently in the network of articles. Second, the homogeneous phenomenon happened more frequently in the network of authors. Third, there are some significant effects in the social influence model and social selection model. Fourth, there are some different results between elementary and high-level exponential random graph models, and elementary models is more likely to have some degenerate problems.