網友推薦系統之研發與建置

Development and Implementation of Recommendation System for Friends

劉德敏(De-Min Liu)

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(2005年)

摘要

網路的蓬勃發展，拉近了人與人的距離，各種現實生活中的行為，全都上搬了網路，如：交友、購物、拍賣等。現今以社群交友為性質的網站比比皆是，在眾多網站的激烈競爭下要如何獲利與經營，是各網站經營者迫切須要思考的議題。交友網站的目的即是交友，因此若能提供給消費者合適的交友推薦，將有助於提升網站的服務品質與經營。概括目前社群交友網站，皆是利用搜尋引擎來提供網友推薦，此項機制的缺點是容易產生過多且不必要的推薦，因此以搜尋引擎來做推薦並不能算一個有效的推薦方法。所以本研究試圖利用新的機制，來產生有效的推薦，採用網站會員間彼此互動強弱的關係，經設計之篩選機制後，提出一套有次序性且符合不同會員個人化需求的推薦結果。
本研究先使用群集分析法，分析出具潛力的會員做交友推薦，再分別以不同的推薦數量做分析，實驗結果顯示20 位網友為最佳推薦數量，並且推薦數量愈少，預測的推薦準確率越高，顯示推薦次序是有效性，因此本研究的推薦法能改善傳統搜尋引擎的詬病，有效的提供推薦的品質，使網站的會員能確切的獲得合適的推薦，可增進網站的營運與獲利。

關鍵字

推薦系統 ； 個人化需求 ； 社群交友網站

並列摘要

The internet tremendously shorten the distances among persons. Most of things have moved to internet in our daily life such as making friends, shopping, selling things. There are too many websites provide the services to make friends in internet. How to promote their services from these competitive market is the most important issue in the friend-oriented website. The goal of the friend-websites is to provide a service platform to make friends including recommending mechanism. To recommend a list that fits individual needs will promote services and increase business of the websites. But most of the websites generate the recommending lists by the seeking engine. However, it is not an efficient way to use the seeking engine to do the recommendations because it usually produces too many and unnecessary lists. This research uses the frequency of the real interactive contacts among members to product recommendations. Our recommend system can provide a personalized and meaningfully ordered list for the new members of websites to increase the speed of group forming.
In this research, we use cluster analysis to choose the potentials to do the friend-recommendation. Then we use the different recommending quantity to maximize the recommending precision and recall ratio. Our experiment shows the 20-member list is the optimum recommending quantity. It also shows the less members we recommend, the higher predicting precision. This experiment shows our recommending list is meaningful and effective. Our recommend system can improve the problems of traditional searching engine in friend-recommendation. Our research provides a qualitative mechanism of friend recommendation. This system gives a personalized list for the new member which can increase the loyalty of customers and promote the level of services in the website.

並列關鍵字

Recommendation System ； Individualized Demand ； Websites of the Friend

參考文獻

【5】Abe, K., Taketa, T., and Nunokawa, H. (2000), “An idea of the agent-based information recommendation system using the statistical information,” 7th International Conference on Parallel and Distributed Systems: Workshops, pp. 143-146.

【6】Adomavicius, G., and Tuzbilin, A. (2001), “Using data mining methods to build customer profiles,” IEEE Computer Society Press, 34(2), pp. 74-82.

【7】Balabanovic, M., and Shoham, Y. (1997), “Fab: Content- Based, Collaborative Recommendation,” Communications of ACM, 40, pp. 66-72.

【10】Goldberg, D., Nichols, D., Oki, B. M., and Terry D. (1992), “Using Collaborative Filtering to Weave an Information Tapestry,” Communication of the ACM, pp. 61-70.

【11】Ha, S., and Park, S. (1998), “Application of Data Mining Tools to Hotel Data Mart on the Intranet for Database Marketing,” Expert Systems With Applications, 15, pp.1-31.