本研究以建置「口碑推薦系統」為目標,探討如何由網路文章對餐廳的討論中,擷取有用的口碑語意,再依據人類對不同類型餐廳的經驗認知與評價標準,訂定出適性化的評價原則,以避免因缺乏彈性而陷入一體適用(one size fit all)的問題;另一方面,推薦的接受方同樣也有適性化的需求,因此須能依使用者喜好,產生接近真實需求的推薦。本研究利用知識本體(ontology)技術,發展所需的知識分類與解題之知識模型,主要的內容包括:(1)釐清此議題的知識源內涵,例如餐廳類型、評價指標、情境、口碑、經驗法則等知識分類;(2)將一般化的知識源建置為領域本體,其內容是由共通性的概念架構及實例共組而成,以利提供其他領域或系統在溝通時的參考標準或術語;(3)以解題需要來發展各知識源之間的邏輯,建立目標導向的任務本體,並以has-a的組合關係建立關聯性,再依據各概念的屬性項目收集現況實例資訊,以做為任務本體的事實知識;(4)最後,發展口碑推薦的語意規則,並以前述的事實知識為基礎,推論隱含性知識。本研究為驗證知識模型效能,選擇以餐飲美食評論為實驗對象,資料來源為所萃取之網路口碑實例資料30例,評估方式分為知識模型評估分析系統彈性及正確性、及使用者推薦正確性,其中系統與人工判讀相同口碑數為90篇,使用者推薦正確率達87.8%,由實驗結果顯示,以知識分類架構為基礎結合本體與推論規則的設計,相較於資料庫系統較有彈性。
This study aims to integrating knowledge taxonomy for a recommender system on restaurant e-WoM, and identify the useful semantic concepts from blogs of restaurant recommendations. To avoid the gap of one-size-fit-all in the recommender systems, this study proposes adaptive formulas with the different experiences and judgments of different types of restaurants for making trustful recommendations. This study bases on the ontology to develop the knowledge model with the field taxonomy and problem-solving concepts. The knowledge model components include: 1) Redefine the knowledge concepts and the knowledge taxonomy of the restaurants fields; 2) a domain ontology to define the common concepts and instance using the is-a relationships for communication in the system; 3) a task ontology to develop the logic based on the problem-solving concepts using the has-a relationships to express the structure of instances and fields knowledge; 4) proposes a set of semantic rules for inference engine reasoning. The result of case shows that the knowledge model enhances the reasoning to support for the users with the restaurant decision-making.