以謠言與謊言理論為基礎之真假評論識別研究;Identifying Deceptive review comments with rumor and lie theories

NCU Institutional Repository > 管理學院 > 企業管理研究所 > 博碩士論文 > Item 987654321/71481

請使用永久網址來引用或連結此文件: http://ir.lib.ncu.edu.tw/handle/987654321/71481

題名:	以謠言與謊言理論為基礎之真假評論識別研究;Identifying Deceptive review comments with rumor and lie theories
作者:	林佳勳;Lin,Jia-Syun
貢獻者:	企業管理學系
關鍵詞:	謠言;謊言;虛假評論;負向評論;文字探勘
日期:	2016-06-29
上傳時間:	2016-10-13 13:07:22 (UTC+8)
出版者:	國立中央大學
摘要:	現今，互聯網普及的時代，人們已習慣頻繁的參與網路上的互動，不僅積極將自身的真實體驗發佈分享，同時也扮演訊息接收者的角色，如此而產生的大量評論已成為多數人購買商品或服務前的參考依據，亦有調查數據顯示，消費者對於線上評論的信任程度為逐年成長的趨勢，其中負向評論更是影響消費者的決策。不幸的是，由於線上評論的快速傳遞特性與巨大影響力，許多組織開始蓄意誇大自身產品或捏造負向評論來攻擊競爭對手，以期從中獲取利益，然而，這樣濫用線上評論的同時，對消費者個人及商業組織皆造成損害。研究顯示，受到網路評論而影響購買意圖最強烈的，即是旅遊業及旅館業，這些評論記錄實用的旅遊資訊及個人經驗，是旅客在出發至陌生景點前的重要參考資訊。故本研究將以芝加哥前二十間知名旅館的負向真實評論及虛假評論為研究對象，包括在六個知名旅遊評論網站上的真實評論，以及由亞馬遜群眾智慧平台Amazon Mechanical Turk10所蒐集的虛假評論。並以謠言及謊言理論為基礎延伸六大屬性：旅館重要屬性字、模糊字、第一人稱代名詞、負面用詞、簡化思考代名詞以及冗詞贅字。運用文字探勘技術結合分類演算法進行分類器訓練，再利用多個單一分類器預測之結果進行分類器整合研究，建構準確且效率兼備的虛假評論識別模型。本研究建構之分類模型結果顯示，利用六大屬性進行資料維度縮減後，不僅讓運算效率提升，同時保持合理的準確性，有效的進行真實與虛假評論之識別，而分類器整合後所得之精確度（Precision）、召回率（Recall）、準確度（Accuracy）及F值（F-measure）四個指標值，也勝過單一分類器之效能，並能有效避免單一分類器可能不適用於其他資料集之風險。 ;Nowadays, people got accustomed to participate interaction frequently on the web in the era of internet. They were not only sharing their real-life experience actively , but also playing the role of the recipient of the message, and thus generated a lot of comments had become references for most people buying products or services. The survey data also showed that consumers trusted online reviews growing year by year, in which deceptive reviews had much more influences on consumer decisions. Unfortunately, due to the rapid transfer and enormous influence were typical of online reviews, many organizations began to deliberately exaggerate their own products or fabricated negative comments to attack competitors in order to derive benefit. However, individual consumers and commercial organizations would cause damage by abusing online reviews. Studies had shown that, it had much more influence by online reviews were tourism and hotel industry. These comments recorded of useful tourist information and personal experience which were important information for travelers in unfamiliar places before departing. This study discussed the negatively truthful review and the deceptive reviews from top twenty famous hotels in Chicago, including the true reviews taking from six famous review sites and the comparison group deceptive reviews on Amazon Mechanical Turk10. On the basis of the rumors and lies theories, the method created six attributes, key words of hotel, vague words、personal pronoun、negative words、pronouns and pleonasm. By using text mining combined classification algorithm to forecast outcome and apply to build models. In this model showed that the mathematical operations not only worked more efficiently but kept the accuracy reasonably, so it could distinguish true or deceptive reviews well. After integrating classifiers, the four indicators for “ Precision”, ”Recall”, “Accuracy”, and “F-measure” had better efficacy than single classifier , also could avoid the risk of unsuitable for other data set.
顯示於類別:	[企業管理研究所] 博碩士論文

文件中的檔案:

檔案	描述	大小	格式	瀏覽次數
index.html		0Kb	HTML	258	檢視/開啟

在NCUIR中所有的資料項目都受到原著作權保護.

社群 sharing

資料載入中.....