運用指數隨機圖模型探討不同性質虛擬社群之社會交換網絡：以批踢踢實業坊為例

Abstract

虛擬社群是當今人們獲取資訊的重要管道，過去文獻指出，不同性質的虛擬社群有其互動模式，並構成獨特的網絡結構，較不宜被均質地討論。虛擬社群被區分為「問答型」、「討論型」、「社會支持型」三種主要的性質，為驗證不同性質的虛擬社群確實展現特有的網絡樣態，本研究針對批踢踢實業坊中各自代表不同社群性質之「C++板」、「健身板」、「狗板」進行探討。 本研究採取不同於過去僅考量成員間單一互動關係的網絡建構方式，而將多元的互動關係完整地納入考量中，試圖建構出更貼近社群真實複雜互動情形的網絡。網絡建構方式的不同的確在一些網絡分析指標量測結果上呈現明顯的差異；此外，指標量測結果也指出不同性質的虛擬社群有著獨特的網絡特徵。因此，本研究進一步運用「指數隨機圖模型」 （Exponential Random Graph Models, ERGMs）觀察不同性質虛擬社群的網絡樣態，尤其關注於社會交換行為的活動，透過對網絡的微觀結構進行統計檢定，而後得知不同性質虛擬社群所擁有的網絡樣態特徵是具有統計意義的。 不同性質的社群因不同程度的討論門檻影響著使用者行為模式，進而形塑獨特的網絡樣態。研究結果顯示：（一）有別於過去文獻的研究結果，「問答型社群」網絡凝聚力事實上是最高的，與「小世界」網絡型態相似，社群中較容易觀察到「直接互惠」與「間接互惠」現象。此結果與本研究更精細的網絡建構方式息息相關；（二）「討論型社群」整體緊密而無明顯的分群情形，本研究提出「萍水相逢」網絡型態來描述之，社群中可能存在「間接互惠」與「擇優連結」的傾向；（三）「社會支持型社群」整體凝聚力雖最低，但其分群結果明顯，代表不同的小團體有其各自感興趣的話題，小團體內部的凝聚力高，但小團體之間仍有較「小世界」網絡型態更為密切的來往。網絡整體與「結構凝聚」網絡型態相似，容易觀察到「間接互惠」的現象，並可能存在「擇優連結」傾向。

Virtual communities have become an important venue for individuals to seek information and find social support. Previous studies indicate that different types of virtual communities have distinct interaction patterns among members, it is therefore inappropriate to treat all virtual communities as the same because of their unique network structures. Specifically, three types of virtual communities have been distinguished: “question and answer”, “discussion” and ” social support”. To examine whether different virtual communities do indeed have different network structure and topologies, three bulletin boards: ”C++”, “MuscleBeach” and “Dog”, were selected from PPT, the largest electronic bulletin board in Taiwan; each represents the aforementioned types of virtual communities. A new approach to network construction that takes into account the full complexity of the interactive nature of the threaded dialogues within the bulletin boards was adopted. Significant differences were found between the proposed and the traditional network construction method on serveral social network measures. It was also found that the three virtual communities did demonstrate very different interaction patterns and network structure. Furthermore, motivated by social exchange theory, we further explore the network typologies in these virtual communities using Exponential Random Graph Models (ERGMs), which allows us to perform statistical test on the differences in interaction patterns. The results show that: (1) Contrary to previous research result, the “question and answer community” demonstrates considerable network cohesivenss and resembles the “small world” typology, mainly due to the more elaborate network construction approach we propose. “Direct reciprocity” and “indirect reciprocity” in this type of community were shown to be statistically more prevalent. (2) The “discussion community” is cohesive and without clearly demarcated network boundaries, hence we propose “chance encounter” typology to portray this kind of community. “Indirect reciprocity” and “preferential attachment” were found to occur frequently in it. (3) The ”social support community” demonstrates very clear network clustering, though considerable denser intearctions were found than in the case of the small world structure. It demonstrates the “structurally cohesive” typology where “indirect reciprocity” and “preferential attachment” were more frequently found.