Please use this identifier to cite or link to this item:
http://tdr.lib.ntu.edu.tw/jspui/handle/123456789/56966
Title: | 依群聚現象建立人際關係網路 Building Relation Network between People by Considering Aggregative Dynamics |
Authors: | Yen-Chun Liu 劉彥均 |
Advisor: | 鄭卜壬(Pu-Jen Cheng) |
Keyword: | 社群網路,鍵結預測,關係強度,貝塔分布,梯度下降法, Social networks,Link prediction,Relation strength,Beta distribution,Gradient descent, |
Publication Year : | 2014 |
Degree: | 碩士 |
Abstract: | 在具有社群網路性質的網路服務中,人與人之間的人際關係網路是一個很重要的資訊,讓我們可以利用這個資訊來增進這些服務的使用性,因此,如何推測人與人之間的關係是很重要的議題。在這篇論文中,我們提出了一個與先前研究不同的方向,我們想要藉由觀察到的人與人之間的互動資訊來推測他們之間的人際關係及它們的強度,在這個前提下,我們提出兩個想法來解決這個問題,第一,如果兩個人是朋友的話,他們會一起出現在同一個物件中很多次,第二,如果兩個人是朋友的話,他們會一起出現在人少的物件中,我們藉由真實資料的觀察來驗證我們的這兩個想法是正確的。根據這兩個想法,我們藉由貝塔分布的概念提出一個非監督式模型,再利用梯度下降法來推測人與人之間的關係。在實驗中,我們提出的方法表現優於其他三個比較方法,此外,我們所推測出來的人際關係網路擁有一般社群網路所擁有的性質。 In the social services, the relation networks between people bring us valuable information to make these services more convenient. Therefore, how to predict the relation between users is an important issue. In this thesis, we propose a new research direction that predicts the hidden relations and their strengths between people according to the observed interaction between them. We have two ideas to solve this problem. First, if two people are friends, they may appear together many times. Second, if two people are friends, they may appear in the object which there are no anyone else. We make observations to verify our ideas first. Then, based on these two ideas, we purpose an unsupervised model by Beta distribution and use gradient descent method to infer the relation between users. In our experiments, our methodology outperforms other three baselines. Moreover, the relation network inferred by us satisfies the constraints for a real social network. |
URI: | http://tdr.lib.ntu.edu.tw/jspui/handle/123456789/56966 |
Fulltext Rights: | 有償授權 |
Appears in Collections: | 資訊工程學系 |
Files in This Item:
File | Size | Format | |
---|---|---|---|
ntu-103-1.pdf Restricted Access | 784.79 kB | Adobe PDF |
Items in DSpace are protected by copyright, with all rights reserved, unless otherwise indicated.