以高價值群顧客為基之聯想式線上社群推薦機制

A Valuable Cluster based Associative Recommender Mechanism for Online Community

林玉國(Yu-Kuo Lin)

指導教授 ： 羅淑娟

國立臺北科技大學/管理學院/工業工程與管理研究所/碩士(2007年)

摘要

現階段有企業編列大量的預算投入網路傳播媒體當中，雖然可以提高企業的知名度，但是卻吸引不了大多數的消費者消費，因為現代的消費者由於積極的交流訊息，因此已經不再完全相信企業傳來的資訊。實際影響消費者購買行為的主要關鍵為消費者之間的口耳相傳，因此，如果能夠將社群網路中較積極且具有價值的會員推薦給其他會員認識進而成為好友，相信藉由同儕團體的關係產生門檻效果、示範效果，會使得其他消費者的消費如滾雪球般的現象，另外，也能夠產生如店員般的會員給予消費者幫助。因此，本研究希望以業界真實資料做分析能夠有效的產生符合個人化推薦，並且使得往後企業透過本研究機制，能夠以較低的營運成本更有效的促進消費者消費。
本研究所提出的模組化推薦機制分為兩階段，第一階段為群集分析，以RFM為顧客特徵進行兩階段式群集分析，分析後找出具有意義的群集數，再將結果加以定義並且命名；在第二階段，本研究應用霍普菲爾網路做為推薦法則，利用第一階段所找到的VIP顧客群與具有潛力顧客群的留言關係做為訓練範例，以過去具有參考價值會員的交友風格做為典範來聯想欲推薦會員的交友風格。實驗結果得到推薦準確率為9.4%，搜全率為56.3%以及F1指標值為16.1%。

關鍵字

RFM模型 ； 群集分析 ； 推薦系統 ； 自組織映射圖網路(Self-Organizing Map，SOM) ； 霍普菲爾網路(Hopfield net，HNN)

並列摘要

Nowadays, a great deal of firms are preparing tremendous budgets into internet media. It could not attract most consumers to purchase their products though it did make companies’ name famous. That is because today’s consumers communicate each other so actively that they don’t completely believe information from these firms. The key to affecting consumers’ purchasing behaviors would be the word-of-mouth advertising among consumers. As a result, if we can introduce those active and valuable members in community to others and make them become friends, consumptions will snowball as a demonstration by the relationship among colleagues. Moreover, these members will give consumers assistance just like sales clerks. Therefore, this research is to produce recommendation of customization by the analysis of actual business data. And then firms can use lower operation budgets to prompt consumers’ purchasing through this research system.
There are two steps of module recommender system in this research offered by our institute. The first step is cluster analyze. We proceed two-stage procedure cluster analyze of SOM+K-means for customers’ characters via RFM. After that we locate meaningful cluster numbers, defining the results and naming them. In the second step, this research applies Hopfield net as a recommendation principle, and uses the messages of VIP and potential customer groups in the first stage as a training demonstration. So we can infer recommended members’ friendship style from valuable members’ in the past. The results of this research are that precision is 9.4%, recall is 56.3% and F1 metric is 16.1%.

並列關鍵字

RFM ； Cluster Analysis ； Recommender System ； SOM ； HNN algorithm

參考文獻

[41] 程士豪，「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」，碩士論文，國立台北科技大學工業工程與管理系，2006。

[45] 劉德敏，「網友推薦系統之研發與建置」，國立台北科技大學學報，第三十九之一期，pp.85-102。

[38] 許智為，「應用巢狀式群集分析方法改善顧客區隔效度之研究」，碩士論文，國立台北科技大學工業工程與管理系，2006。

[1] Adomavicius G,Tuzhilin A., "Toward the Next Generation of Recommender Systems:A survey of the State-of-the-Art and Possible Extensions," IEEE Transaction on Knowledge and Data Engineering, 2005, pp.734-749.

[2] Adomavicius, G., and Tuzbilin, A., "Using data mining methods to build customer profiles," IEEE Computer Society Press, 34(2), 2001, pp.74-82.