個人化網友推薦系統之建立與效度改善 -- 應用於某社群網站

Development and Effect Evaluation of Personal Friendship Recommendation System -- A Case Study of Community Website

程士豪(Shiu-Hao Cheng)

指導教授 ： 羅淑娟

國立臺北科技大學/管理學院/工業工程與管理系所/碩士(2006年)

https://doi.org/10.6841/NTUT.2006.00113

摘要

Internet的快速發展，使得人類的許多行為並非真實面對面來達成。坊間許多功能型的網站的辟立，足以證明這點。如專門討論區社群網站、購物網站、交友網站等等。對於網站的經營，在網站使用者與網站經營者的角度，都是希望達到雙贏的地步。所以在眾多的網站競爭之下，要讓網站的會員忠誠度高並進一步產生獲利，其實是網站經營者最需要深入的議題。當然對於網站使用者，使用性越方便往往是最為上策的。環顧現今的社群網站，最常見的是利用使用者勾選網站所提供之選項，來搜尋配對網友的推薦。此機制已行之多年，對於網站使用者來說，是容易產生過多的推薦，這對於較為被動的網站使用者，並不能算是種有效的推薦機制。本研究提供新機制，擷取網站使用者的歷史資料，採用會員間的留言來分析互動關係，經設計之篩選後，產生以個人會員為主並有次序的推薦名單。
本研究先使用群集分析法，過濾出具有潛力的會員作為本網站推薦對象。再分別以5~30位，六種不同的推薦數量來做精確率(precision)、搜全率(recall)、F1值的分析，結果顯示以20位的網友為較優推薦人數。並且把實驗結果與文獻做比較。顯示本研究方法確實能有效提高精確率，並有效的提供推薦品質。

關鍵字

推薦系統 ； 個人化 ； 社群網站

並列摘要

After the development of Internet, human’s comportment or behavior don’t need to be face to face anymore. And that’s why those web-sides all over the internet world, such as online community, shopping web and chat room. About how to run the Internet business, both web-visitor and web-operator wants to get the benefits by Internet web-side. To be surviving in this competition, web-operator needs to focus and learn how to raise the loyalty of web-visitor. And the convenient of web-side could be the most important to web-visitor. To make down the choice of web-side by items selections on web-side is the common way for web-visitor, which could accord the web-visitor’s need. This mechanism already use for years, web-visitor will get excrescent suggestions more then they need, which is not a convenient mechanism to passive web-visitor. The thesis provides a new mechanism. According to history information of web-visitor, adopt the massages and conversation to analyze the interaction. After design and sieve, produce the recommend list by order.
This thesis use cluster analysis, sieve potential web-visitor and recommend to web-side. 5-30 people, use six different recommend amend to analyze precision, recall, F1. And the result show 20 web-visitors would be the best recommend amend. And also compare with experiment result and literature. Prove this organon could really raise precision and also provide high quality.

並列關鍵字

Recommendation System ； Personalization ； Online Community

參考文獻

[3] Chow, C.K. and Liu, C.N, "Approximating discrete probability distributions with dependence trees," IEEE Transactions on Information Theory, 14, 1968, 462-467.

[5] Everett M. Rogers , Diffusion of Innovations. New York: Free Press, c1995 ,4th ed.

[7] Goldberg, D., Nichols, D., Oki, B. M., and Terry D, "Using Collaborative Filtering to Weave an Information Tapestry, " Communication of the ACM, 1992, pp. 61-70.

[8] Herlocker, J. L., Konstan, J. A., and Riedl, J, "Explaining Collaborative Filtering Recommendations," ACM 2000 Conference on Computer Supported Cooperative Work.

[9] J. Cheng, D.A. Bell, W. Liu, "An Algorithm for Bayesian Belief Network Construction from Data, " In Proc. of AI & START’97 , 1997, 83-90.