Finding Customer Opinions based on User-given Aspect

NCUIR > School of Management at National Central University > Graduate Institute of Information Management > Electronic Thesis & Dissertation > Item 987654321/65652

Please use this identifier to cite or link to this item: http://ir.lib.ncu.edu.tw/handle/987654321/65652

Title:	Finding Customer Opinions based on User-given Aspect
Authors:	黃于珊;Huang,Yu-Shan
Contributors:	資訊管理學系
Keywords:	意見探勘;NGD;WordNet;關鍵字註解;構面分析;Opinion Mining;NGD;WordNet::Similarity;Keyword Annotation;Aspect Analysis
Date:	2014-07-14
Issue Date:	2014-10-15 17:07:25 (UTC+8)
Publisher:	國立中央大學
Abstract:	由於網際網路的快速發展，許多服務與購物網站累積了許多評論資訊。為了提供給消費者更多資訊使得使用者能更快速找到自己需要的評論，這些電子商務網站會利用事先定義好的產品構面將使用者的評論進行分類。以Hotels.com為例，他們事先將評論分為服務、清潔度、舒適度、整體外觀等四個構面，並給予每個構面分數。然而，依據各網站自我事先的定義，無法針對每位使用者個別的特殊需求，讓使用者找到自己所關心的相關評論資訊。因此本研究提出新的評論分析方法，能夠利用使用者能動態給予的產品構面找到相關的顧客評論。本研究包含四個主要步驟：前處理(pre-processing)、註解(annotating)、配對(matching)與排列(ranking)。而在註解過程中執行三個不同的方法找到與使用者提供的構面相關的關鍵字。而在配對過程中，利用Google與WordNet進行關鍵字的相似度計算，並利用四個不同的鏈結方法計算構面與句子的相似度。最後利用每個句子與構面的相似度排列評論句子提供給使用者。;As the Internet grows day by day, customer reviews in e-commerce websites are also increasing every day. To provide users more information, these commercial websites usually would summarize users’ opinions in reviews according to some predefined aspects. For example, Hotels.com extract scores from customer reviews in four aspects, including service, cleanliness, comfort and condition. However, the weakness of this predefined analysis approach is that users cannot find opinions on the aspects that have not been considered beforehand by the websites. To solve this problem, we propose a new approach to generate a ranking of related opinions based on user’s dynamically given aspect. The proposed approach involves 4 phase named pre-processing, annotating, matching and ranking. In the annotating phase, 3 different annotating methods are used to annotated related keywords based on user-given aspect. In the matching phase, Google similarity and WordNet similarity, are first used to compute the similarity between keywords and then the similarity between aspect and each sentences is computed by 5 different linkage methods. Finally, we rank sentences based on their similarities with aspect.
Appears in Collections:	[Graduate Institute of Information Management] Electronic Thesis & Dissertation

Files in This Item:

File	Description	Size	Format
index.html		0Kb	HTML	508	View/Open

社群 sharing

Loading...