|Abstract: ||在Web 2.0的快速發展與電子商務盛行的環境下，電子商務平台提供消費者一個完整的商品購買循環：購買前搜尋商品資訊與消費者意見、線上消費產生商品訂單與結帳支付、售後服務與反饋。消費者能以線上評論的方式，分享使用經驗及意見回饋，亦能自該社群獲取產品、服務資訊以作為產生最終消費行為之參考，經此互動所產生的訊息內容形成蘊含強大電子口碑的用戶生成內容資料庫，其中具有較高的有益性的評論不僅是影響消費者產生最終購買決定之參考，更成為影響對於商家銷售在產品、服務獲益之重要因素。每天產生龐大的評論內容，雖然可以充足地提供給消費者富有價值的訊息，卻也衍伸出資訊過載的問題，因此許多的商家加入排序機制與條件篩選幫助消費者在有限時間下選擇性地閱讀評論內容，並減少消費者在資訊搜尋程序所需之成本。本研究旨在以長期性地觀察線上評論因時間、條件篩選與排序機制後所陳列位置的移動狀況，從消費者的角度提出影響可視性之因子，探討其對於評論有益性之影響，並建構有效線上評論有益性的預測模型。透過Python 撰寫網路爬蟲程式，針對目標產品之線上評論內容及其於程式運行當天所在之網頁位置資訊，並利用自然語言處理產生自變數，以評論品質(Review Quality)、文本特徵(Text Characteristic)與評論可視性(Review Visibility)三個分類構面進行變數分類，採用線性迴歸(Linear Regression)、迴歸樹(Regression Tree)、支援向量機迴歸(Support Vector Regression)及最近鄰居法(k-Nearest Neighbors Algorithm)四種資料探勘技術進行線上評論有益性分析及預測。透過研究結果發現，以迴歸樹建立之有益性預測模型效能最好，並且加入評論可視性的變因便能顯著地提升評論有益性之預測模型效能，而對於商品評論及旅館評論兩者將有不同的最佳變數組合，藉此幫助消費者或是電子商務業者找出最有幫助的評論。|
Under the circumstances of rapid growing Web 2.0 and prevailing electronic commerce, e-commerce platforms offer consumers a complete product purchasing cycle, which includes searching for product information and consumer opinions before purchasing, completing an online purchase and paying for the order, enjoying post-purchase services and giving feedbacks. Consumers can share their own using experiences and opinions by posting online reviews. They can also obtain product or service information from the social networking site and take it as a reference when making final purchase decision. The review contents produced through interaction form a user generated content database which contains powerful electronic word-of-mouth. The reviews that have the highest helpfulness not only impact consumers' final purchase decision, but also are the factors that can affect a business's sales on product and service. There is an enormous amount of reviews generated every day. Although these reviews provide sufficient and valuable information, they derive the problem of information overload. Therefore, many businesses use sorting and filtering mechanism to help consumers reading the reviews selectively and reduce the costs of their information searching process.The main purpose of this research is to do a long-term observation on the moving situation of arrangement of online review positions based on time, sorting and filtering mechanism. We identified some factors that might affect the visibility from consumers' perspective, explored their influences on review helpfulness and constructed an effective forecasting model of online review helpfulness. We implement a web crawling program in Python which crawled the online reviews of target product as well as its web page location, and used natural language processing to generate independent variables. We classified the variables by three aspects, which are review quality, text characteristic, and review visibility. We also adopted linear regression, regression tree, support vector regression, and k-nearest neighbors algorithm, the four data mining technologies to do analysis and make prediction of online review helpfulness. There are some results found in this research. The forecasting model of helpfulness constructed with regression tree has the best effectiveness. Moreover, adding variable of review visibility can significantly increase the effectiveness of review helpfulness forecasting model. There are also different combinations of variables for product reviews and hotel reviews, which can help consumers or e-commerce businesses find the most helpful reviews.