以本體論導入個人化行動情資於成對推薦之研究

The Ontology Approach for Personalized Pair-Recommendation--Using Mobile Message as Example

蒲琮仁(Tsung-jen Pu)

指導教授 ： 李麗華

朝陽科技大學/資訊學院/資訊管理系碩士在職專班/碩士(2011年)

摘要

近年來由於無線科技的逐漸進步與行動裝置應用上的普及化，企業M化的概念和議題也隨之興盛，企業M化不僅提升企業內部的營運效率，同時成為企業間彼此競爭之重要指標。對行動使用者來說，使用者可不受時空之限制，隨時(Anytime)、隨地(Anywhere)接收服務提供者(Service Provider)所傳送的行動情資(Mobile Message)，相對地，此便捷之方式將形成許多問題，例如連續的發送行動情資，可能導致使用者接收的情資數量供過於求，且未符合使用者需求而淪為不必要的情資，甚至可能造成使用者對系統的反感與不信任感。
為解決上述問題，周裕健學者於過去提出一個以本體論技術建構個人化行動情資之推薦機制(Personalized Mobile-Message Recommendation, PMR)，運用本體論建立情資範例本體論及使用者相關資料。為分析本體論資料間是否相似，以本體論比對(Ontology Matching)技術進行情資類別間的差異性比對，分別設計三種本體論結構相似性比對方式(1)廣度優先比對(Breadth-First-Matching, BFM)、(2)深度優先比對(Dept-First-Matching, DFM)與(3)節點索引比對(Node-Index-Matching, NIM)，並評估其比對效能，經實驗證明，NIM具有較佳的比對效能。針對產生推薦列表的部分，以多數相似門檻值(Most Similarity Threshold, MST)的方式擷取可推薦項目，此門檻值以單純的多數決概念定義，換句話說，使用者對行動情資的瀏覽數量須達情資範例本體論節點總數的一半，故系統無法針對行動情資瀏覽量較低的使用者產生推薦。
為改善上述問題，本研究提出以本體論導入個人化行動情資之成對推薦(Personalized Pair-Recommendation, PPR)，成對推薦之概念來自搭售(Bundling)策略，PPR旨在以搭配之方式推薦行動情資給予使用者，不以銷售為最終目的，除此之外，PPR以個人化為主，分別運用內容導向式推薦(Content-based Recommendation, CB)與協同過濾式推薦(Collaborative Filtering Recommendation, CF)產生成對推薦給予使用者。
透過實驗證明，本研究提出的兩種PPR推薦方法，不論在推薦成功率或F1指標的效能表現，皆優於其它推薦方法，故本研究採用混合式推薦，依據使用者對行動情資的瀏覽特性及傾向，分別使用兩種推薦方法產生推薦。

關鍵字

成對推薦 ； 搭售 ； 行動情資 ； 推薦系統 ； 類神經網路 ； 本體論 ； 自適應共振理論

並列摘要

Recently, due to the progress of wireless technology and the popularity of mobile device applications, the concept and topic of enterprise mobilization are frequently discussed. Enterprise mobilization not only can enhance the internal operation efficiency of an enterprise, but also can be an important indicator for competing with other enterprises in the marketplace.
The mobile user is able to receive the mobile message anytime and anywhere without the restriction of space-time, nevertheless, such a convenient way also brings some problems, for example, in the condition of continuous passing mobile message, the number of mobile messages received by the mobile user will be too much or redundant. In a more serious consequence, it may cause the mobile user’s dissatisfaction and distrust to the system.
The researcher Chou (2009) proposed a Personalized Mobile-Message Recommendation (PMR) module in order to solve the problem that is mentioned above. The PMR system applies ontology to construct the mobile message template and mobile user-relevant information. In order to analyze the similarity between mobile message template and mobile user-relevant information, the ontology matching will be used to compare the difference of the mobile message. This study designs three different ontology structure similarity matching methods, such as (1) Breadth-First-Matching (BFM), (2) Dept-First-Matching (DFM), and (3) Node-Index-Matching (NIM) methods. Among these 3 methods, the NIM has better matching performance. For the recommendation phase, PMR system produces the recommendation list based on Most Similarity Threshold (MST) to extract the appropriate mobile message to the mobile user. MST means the browsing number of the mobile message has to exceed a half of the total number of the mobile message, therefore, PMR module cannot generate recommendation for the user that has less browsing number of mobile message.
In order to improve the problem that is mentioned above, this study proposes a Personalized Pair-Recommendation (PPR) for the mobile message based on ontology approach. The concept of Pair-Recommendation springs from the bundling which is a type of marketing strategy. However, the purpose of pair-recommendation is to recommend related and preferred paired message rather than focus on the sale’s rate. The proposed PPR will analyze the mobile user’s preference according to the personalized ontology of user’s preference. The proper mobile messages are pair-recommended to various users by using the Content-based (CB) method or Collaborative Filtering (CF) method.
According to the results of the experiment, the proposed Personalized Pair-Recommendation (PPR) has better outcome in terms of successful rate (SR) and F1-Measure. Consequently, we adopt the hybrid recommendation method to recommend the mobile message and to achieve the upersonalized recommendation.

並列關鍵字

Mobile Message ； Bundling ； Recommendation ； Recommendation System (RS) ； Artificial Neural Network (ANN) ； Adaptive Resonance Theory (ART) ； Ontology

參考文獻

[1] 周裕健 (2009)，運用本體建置個人化行動情資之推薦，碩士論文，朝陽科技大學資訊管理系，臺中。

[5] 詹尚驥 (2008)，個人化文章自動推薦-以學術文獻部落格為例，碩士論文，朝陽科技大學資訊管理系，台中。

[10] W. J. Adams and J. L.Yellen (1976), “Commodity Bundling and the Burden of Monopoly,” The Quarterly Journal Economics, Vol. 90, No. 3, pp. 475-498.

[11] A. Albadvi and M. Shahbazi (2009), “A Hybrid Recommendation Technique Based on Product Category Attributes,” Expert Systems with Applications, Vol. 36, No. 9, pp. 11480-11488.

[13] A. Ansari, S. Essegaier, and R. Kohli (2000), “Internet Recommendation Systems,” Journal of Marketing Research, Vol. 37, No. 3, pp. 363-375.