- 學位論文
基於共同話題之潛在朋友推薦系統
A Recommender System of Potential Friends Based on Common Topics
摘要
近年來由於虛擬社群的蓬勃發展，人們正想著該如何藉此增加彼此的互動，於是藉此互動而衍生成社交網站。然而社交網站的興起，代表著另外一種生活型態的轉變，由此轉變，社交網站已經儼然成為一種熱門的社交生活方式。過去的研究當中，相當多數是針對潛在的行為進行分析統計，但是對於藉由文字當中找出使用者之間的關係的研究較為缺乏，對於使用共同話題的找尋朋友方式也較為少見，並未將使用社會網路分析中的相互關係權重加入考量，若單使用社會網路分析權重，只找出重要關鍵節點，所考慮的因素較缺乏。 本研究將提出一個新的方式來改善現有的朋友推薦系統，其中包括三種模組：(1)社會網路分析模組、(2)文字探勘模組及(3)推薦模組，透過社會網路分析、文字探勘及協同過濾方式，計算出發表者在社會網路中的中介中心性，並加上文字探勘方式與控制詞彙所評估出的詞彙頻率並找出主題，透過推薦模組計算相似度，並藉此相似度來預測該使用者的評估值，最後推薦出同一主題下，找出其值相似程度較高的使用者，達到使用共同話題推薦朋友之目的。
並列摘要
The social network phenomenon is with such strong momentum that it has resulted in the change of lifestyle. In this new lifestyle, social network sites have created new ways of socialization and interaction. Traditionally, researchers have often used statistical methods to analyze online social behaviors, but the research can lack authenticity since the social relationships are represented by the networking among the online user postings with common topics. The weighting of social network analysis is usually not taken into consideration. With such traditional methods of social network analysis, the key nodes found and networking factors identified tend to be oversimplified. This research proposes a new method to improve the existing friends’ recommender systems of the social network sites. The recommender system consists of three methods: (1) social network analysis model, (2) text mining model, and (3) recommendation model. In such system, we calculate betweenness centrality and term frequency of online social network postings, and the estimated value is generated by the collaborative filtering model. Finally, we recommend the users with the highest degree of similarity on the same topic to achieve the purpose of our research.