運用本體論建置個人化行動情資之推薦

Using Ontology for Personalized Mobile-Message Recommendation

周裕健(Yu-Chien Chou)

指導教授 ： 李麗華

朝陽科技大學/資訊學院/資訊管理系/碩士(2009年)

摘要

在電子商務環境中，商家在傳送情資時(Message)，內容多數以文字方式呈現，其傳送方式也從原來的網際網路逐漸改以行動方式傳送，因此，帶給商家及消費者莫大的效益，例如，商家可以節省成本，而消費者可隨時、隨地接收行動情資(Mobile Message)，但也因為如此方便，行動情資在不斷的發送下，可能會過於氾濫且無法符合使用者興趣或需求。此外，同樣商品卻有著不同名稱的問題亦可能造成情資推薦不準確。
有鑑於此，本研究提出一個以本體論技術來建構個人化行動情資之推薦機制(PMR)。本研究以本體論(Ontology)技術建置行動情資本體論範本及使用者相關資料，然而為了分析不同的本體論資料間是否相似，因此本研究提出本體論比對技術(Ontology Matching)在項目間進行差異性比對，不過由於在比對過程中本體論結構上的議題仍須探討與解決，因此本研究設計三種不同的本體論結構相似性比對方法進行比對及效能評估，這三種比對方法分別為廣度優先比對(Breadth-First-Matching, BFM)、深度優先比對(Depth-First-Matching, DFM)及節點索引比對(Node-Index-Matching, NIM)。在比對過程中，效能評估主要以比對時所產生的比對次數(MT)為主要評估數據，在本研究提出的三種比對方法實驗結果發現，以節點索引比對方法(NIM)並結合排除空節點(NE)及比對後位置的比對(AIM)的效能最佳，因此較適用於行動環境之情資推薦。
依據上述結果，本研究運用節點索引比對(NIM)方法進行後續之行動情資推薦、效能評估及可推薦清單之實驗。推薦及效能的評估主要是以協同過濾推播模式(CFPM)進行實驗，即依據目標使用者找尋相似之群組使用者，由比對(match)來源端使用者的可推薦清單，產出可推薦清單後再加以推薦，藉此建構並完成個人化行動情資推薦。

關鍵字

點對點 ； 行動情資 ； 推薦系統 ； 本體論 ； 本體論比對

並列摘要

It’s observed that the message delivering and product recommendation have moved from Internet to mobile device. Indeed, the customer is highly beneficial because the customer can receive mobile message anytime and anywhere. However, the unsoliciting messages and the synonym terms could cause problems for applying mobile-message recommendation.

To handle these problems, this paper purposes a Personalized Mobile-Message Recommendation (PMR) system. The PMR system applies ontology to construct the user preference profile and the message template which contains the product or service information. To analyze the similarity between ontologies, the ontology-structure matching is studied because it can effectively search the portion of similar or dissimilar among ontology structures and, therefore, the matching process is faster. To speed up the ontology-structure matching performance, this paper proposes three methods for ontology-structure matching and these three methods are (1) Breadth-First-Matching (BFM), (2) Depth-First-Matching (DFM), and (3) Node-Indexing-Matching (NIM). During the matching process, the Match Total (MT), Match Error (ME), and Match Success (MS) are recorded and the performance is evaluated by using MT, ME, and MS. The experimental results show that the proposed methods are practical for mobile message recommendation. Among these three methods, the Node-Index-Matching (NIM) spends the least node-comparing count and it processes faster which will be promising for mobile message recommendation.

For the recommendation part, similar users are clustered into a group based on the ontology tree of target user. The NIM is used for matching and comparing the ontologies. The recommendation is generated from the leaf-node of similar users’ ontology. The recommendation results are evaluated using precision, recall, F1 and the overall coverage. The outcome shows that the proposed PMR system is applicable for the personalized mobile-message recommendation.

並列關鍵字

Peer-to-Peer ； Mobile Message ； Recommendation Systems ； Ontology Matching ； Ontology

參考文獻

[6] 鄒孟訓(2006)，含可調整及分散式比對方法的知識本體校準系統，國立中央大學資訊工程系碩士論文，桃園。

[11] Dominik Benz, Karen Tso, and Lars Schmidt Thieme, 2007, “Supporting collaborative hierarchical classification bookmarks as an example,” Computer Networks, vol. 51, no. 16, pp. 4574-4585.

[14] Chris Cornelis, Jie Lu, Xuetao Guo, and Guanquang Zhang, 2007, “One-and-only item recommendation with fuzzy logic techniques,” Information Sciences, vol. 177, no. 22, pp. 4906-4921.

[17] Yoon Ho Cho and Jae Kyeong Kimb, 2004, “Application of web usage mining and product taxonomy to collaborative recommendations in e-commerce,” Expert Systems with Applications, vol. 26, no. 2, pp. 233-246.

[20] Hong-Hai Do and Erhard Rahm, 2002, “COMA - A system for flexible combination of schema matching approaches,” Proceedings of the 28th VLDB Conference, Hong Kong, China.