訊息影響力預測：使用Facebook資料為例;Predict Influence of Posts：Using Data from Facebook

NCU Institutional Repository > 管理學院 > 資訊管理研究所 > 博碩士論文 > Item 987654321/54923

jsp.display-item.identifier=請使用永久網址來引用或連結此文件: http://ir.lib.ncu.edu.tw/handle/987654321/54923

题名:	訊息影響力預測：使用Facebook資料為例;Predict Influence of Posts：Using Data from Facebook
作者:	鄭如筠;Cheng,Ju-Yun
贡献者:	資訊管理研究所
关键词:	資料探勘、社群網路、分類;Data mining、Social network、Classification
日期:	2012-06-30
上传时间:	2012-09-11 19:12:47 (UTC+8)
出版者:	國立中央大學
摘要:	社群網站已成為近年來人們最常使用的網站之一，人們可以在社群網站上透過個人檔案分享或是和其他使用者的聯繫與溝通展現自我。Facebook是眾多社群網站中最受歡迎且使用者最多的網站，許多公司為了可以直接與網路上的客戶聯繫，皆已在Facebook架設公司專屬的粉絲專頁或社團。本論文蒐集了Facebook的資料集並依此建立混和模型以預測訊息在經過一段給定時間後其影響力程度，影響力程度的預測主要根據該訊息內容、時間特性及作者特性。本論文之混和模型利用投票的方式整合了五種分類器的分類結果，包含類神經網路、決策樹、羅吉斯回歸、貝式分類及支援向量機。過去的研究在預測訊息重要性時只考慮訊息被使用者存取的次數，忽略了存取該訊息的使用者在社群網站中之個別影響力。本篇論文和過去研究最大的相異之處即在於我們假設每個使用者皆有不同的權重，反應其在社群網路中的個別影響力，因此Facebook上的訊息其影響力即為針對該訊息按”讚”的使用者的權重加總。本篇論文的實驗採用資料探勘工具Clementine執行十折交叉驗證。實驗結果顯示本篇論文採用的混和模型其預測表現皆優於上述的五個分類器，且在本篇論文中提出的預測因子也具有相當的顯著性。實驗結果也說明了在預測訊息的影響力時，若同時考量使用者的個別影響力，其預測結果會相較只考量訊息被存取的次數而忽略使用者的個別影響力時準確。Social networking web sites (SNWs) have become one of the several main sites where people spend most of their time. People can present themselves on their individual profiles, make links to other users, and communicate with them on SNWs. Facebook is one of the most popular media of SNWs and becomes the top most-trafficked website in the world. In order to contact with on-line customers, many corporations have their own page on Facebook. In this study, we focus on Facebook and build an ensemble model to predict the influence of posts in the future based on content features, temporal features, and authorial features. The ensemble model integrates results from Neural Network, Decision Tree (C5.0), Logistic Regression, Naive Bayes, and Support Vector Machines (SVM) by voting method. Different from previous research in predicting influence of posts on SNWs which only consider their access counts and neglect different influence of individual user, this work assumes that each user is associated with a weight to reflect his influence in social network and the influence of a post on Facebook is defined as the weighted sum of the influence of the users who clicked “like. Our experiments are executed by the data mining tool, Clementine, and performed by a 10-fold cross-validation. Experiment results show that the predicting performance of our ensemble model outperforms each individual classifier and the features we propose can significantly improve the prediction of posts' influence. The results also show that our model, which considers different weights of users, can achieve higher accuracy than traditional model, which treats all users the same, in predicting influence of posts.
显示于类别:	[資訊管理研究所] 博碩士論文

文件中的档案:

档案	描述	大小	格式	浏览次数
index.html		0Kb	HTML	598	检视/开启

在NCUIR中所有的数据项都受到原著作权保护.

社群 sharing

数据加载中.....