| 摘要: | Web2.0時代來臨和電子商務的蓬勃發展，讓越來越多人在網路上分享與表達個人對於產品與服務之使用意見。隨著網路拍賣也逐漸普及化，由於通路多元與成本低廉，商家如雨後春筍在拍賣平台上開店。使用者有意圖進行購買行為時，常常會面臨到五花八門的選擇，而需要參考其他買家對產品及服務的評價意見，來輔助進行購買決策。 目前台灣許多大型網路拍賣平台 (例如: 雅虎奇摩拍賣、露天拍賣) 只呈現最基本的二元評價機制，加總使用者給予每一筆交易記錄的評價 (優良評價為+1；差勁評價為-1；普通評價為0)，計算賣家總評價。然而賣家總評價並不能拿來當作比較兩個賣家優劣的評斷標準，也無法透過總評價了解賣家在商品品質、服務態度、送貨速度方面的表現，而這些賣家特性可以幫助使用者對沒有交易過的賣家有更深度地了解。即使使用者可以從賣家的評價記錄看到所有買家的意見，但是在大量的交易筆數中，若以人工處理方式設法獲取具有參考價值的資訊，勢必會耗費龐大的搜尋成本。 因此，本研究擷取網路拍賣平台上的真實交易紀錄，利用意見探勘 (Opinion mining) 的技術，過濾掉沒有參考價值的罐頭意見，根據詞庫自動標示並分類具有意義的特徵 (Feature) ，將以往只能分辨出正評與負評的意見透過情感分析(Sentimental analysis)擴展成情緒量表，進行意見數值化，並計算四個面向 (商品品質、服務態度、送貨速度、整體評價) 的分數，產生能呈現賣家完整特性的使用者輪廓 (User Profile) ，以提供買家進行購物決策之參考指標。除此之外，本研究考量到每筆評價意見之買家的可信度 (Credibility)會因為意見相似度而降低其權重，透過演算法計算並與相關研究整合，改良既有的評價機制。 Online auctions have become immensely popular and created massive cash turnover in recent years. However, for a user intent on purchasing an item from an auction site, selecting an appropriate seller from the numerous choices is not an easy task. The emergence of Internet has constructed a space for users to freely express opinions and exchange experiences regarding products, services, and any public issues. Buyers in online auctions write feedback opinions to the sellers from whom they have bought the items. Other buyer read these opinions to help them determine which seller and item to bid for. In general, many large auction sites in Taiwan (for example: Yahoo auction, Ruten auction) showed only the most basic binary evaluation system, let user rate the seller in three levels, ‘‘good is +1’’, ‘‘bad is -1’’, or ‘‘neither is 0,’’ and calculate the total evaluation of the seller. However, the total evaluation can’t be the judge of merits to two different sellers, and let buyers understand the seller of the product quality, service, and delivery speed. In this research, we aim at helping buyers to make decision in online auction by creating a seller’s user profile. First, we use opinion mining technology to filter out comments, automatically tagged and classified meaningful features under the thesaurus. Second, calculating the four dimensions (product quality, service, delivery speed, and overall evaluation) scores by emotion scales extended from sentimental analysis in the past. Finally, we also take into account the credibility of the buyers that decreases the weight because of the similarity of their opinions. The credibility calculated by algorithms integrated with related research, improving the existing evaluation mechanisms. |
