網路口碑為消費者透過部落格、討論區、電子佈告欄及電子郵件等虛擬平台,傳播對於某一產品或服務之評論。隨著網際網路的快速發展,網路口碑資訊倍增,使用者需花費更多時間過濾無關的資訊,過多的資訊提高消費者閱讀及判斷的困難度,也因為網路的匿名特性使口碑資訊較易受到質疑。本研究針對上述問題,嘗試(1)擴大資料檢索範圍,累積文件數量提高語意分析的可信度;(2)發展語意解析機制,分析口碑之內涵以利推薦。主要分為三項設計,首先,搜尋網路文件萃取本文資料,以利文件語意解析;再者,解析餐飲評鑑之評鑑項目,籍由評鑑項目擷取語意特徵,及量化特徵資訊;最後,針對解析之評鑑項目依知識工程方法進行知識塑模,擬定產品認知及口碑分析特徵之描述邏輯與語意規則,建立推薦之語意模型,配合產品推薦決策推導口碑推薦之結果。 本研究為驗證系統效能,選擇以餐飲美食評論為實驗對象,資料來源媒介為部落格平台,評估方式分為知識模型評估分析系統彈性及正確性、及文章分析正確性,挑選10家餐廳共取得有效文章數為387篇,其中系統與人工判讀相同文章數為341篇,文章正確率達88.11%;再者,由實驗結果顯示,知識本體結合推論規則的設計,可獲得正確口碑推薦,相較於資料庫系統較有彈性。
Electronic Word-of-Mouth (eWOM) is disseminating the comments of products or services through virtual platform by consumers. The virtual platform are various, such as blogs, discussion boards, electronic bulletin boards and e-mail etc. More information let users more difficult to read articles and make decision. Users need to spend more time to filter irrelevant information. This study tries to (1) Expand the scope of information retrieval and accumulate a large number of documents to improve the credibility of semantic analysis. (2) Develop mechanisms for the semantic analysis and analyzing eWOM to recommend. There are three main designs. First, search the web documents and extract the information in the articles, in order to facilitate semantic analysis of documents. Second, we analyze rating items about restaurant ratings in order to capture the semantic characteristics and converted the characteristics into quantities. Finally, this study utilizes knowledge-intensive approaches to represent restaurant recommendation models into ontologies. Develop description logic and semantic rules for cognitive of products and analysis of documents. Users get feedback through the knowledge of inferring. This study verifies the accuracy of restaurant recommendation systems. We collected the articles of blog as experiment. Estimative methods are included that analyze flexibility and accuracy for this system, and analyze accuracy of articles. An ontology-based model with rule can capture accurate recommendations of eWOM and more flexible than RMDB.