Recommendation of Similar Users in Social Resource Sharing Websites

[[abstract]]透過Web2 0大眾分類法(Folksonomy)的標籤功能，使得網路上的社會資源分享網站變成一個知識管理與分享的重要平台。此社會資源分享網站剛開始僅提供書籤收藏與分類管理功能，近幾年增加了「朋友」的社交網路功能，有些還將朋友再細分為「跟隨我的人(Follower)」與「我追隨的好友(Following)」兩個面向。這樣的改變使得社會資源分享網站不再是單純提供使用者管理或分享收藏書籤，而是一個具有社交功能的知識分享網路。 加入社交網路功能之後，提供了設計一個協同推薦系統的新方向。因此，本研究以使用者的標籤資訊為基礎，提出使用者相似度計算指標，並建立協同推薦模型，依據以下兩個因素探討推薦模型的效果是否造成差異：不同好友面向與不同好友階層兩個因素。 推薦方法包含下列主要步驟：(1)建立項目輪廓與使用者輪廓檔、(2)從第一層朋友網路中找出多位相似朋友、(3)再從第一層相似朋友的好友中，找出尚未與目標使用者建立直接好友關係的多位相似使用者、(4)分別推薦前述各層的多位相似使用者給目標使用者。 本研究以Web2 0社會標籤網站Diigo為例，根據該網站所收集到的使用者為資料來源，進行推薦相似使用者給目標使用者，以展示推薦流程與評估效果。調查三十位實驗者，實驗結果發現：(1)本研究之推薦模型之推薦效果優於熱門推薦；(2)本研究推薦模型的計算指標得到的推薦順序是恰當的；(3)使用者自行與Following建立關係，因此，由Following面向推薦相似使用者，相較於從使用者被動接受關係的Follower面向推薦使用者，前者之效果明顯優於後者。[[abstract]]The label function of Web 2 0 Folksonomy enables social resource sharing websites to become a crucial platform for managing and sharing knowledge over the Internet Initially social resource sharing websites simply provided bookmark collection and classification functions the social networking function of “friends” was added recently A number of websites even classify friends further into “follower” or “following” categories This change makes social resource sharing websites knowledge sharing websites with socializing functions instead of simply enabling users to manage or share collected bookmarks A new collaborative recommendation model was designed after the social networking function was added Therefore this study proposes a user similarity calculation index based on the user label information and establishes a collaborative recommendation model Subsequently we explore whether the two factors that is friend dimensions and friend layers cause a differing recommendation model effect The recommendation method includes the following steps: (1) Establish item and user profile; (2) locate similar friends in the first layer of friends in the network; (3) locate similar users that have not established direct friendships with the target users from the first layer of similar friends; and (4) recommend similar users from the previous layers to the target users This study uses the Web 2 0 social bookmarking website Diigo as an example and presents the recommendation procedure and the effects of recommending similar users to target users according to the user information collected by the website The results of surveying 30 participants indicated the following: (1) The effect of the recommendation model developed in this study is superior to that of popularity recommendations; (2) the recommendation sequence obtained by the index calculation of the recommendation model developed in this study is appropriate; and (3) users can establish relationships with their friends in the “following” category Therefore the effect of recommending similar users from the “following” category is superior to users passively accepting users recommended from the “follower” categor