Using Multi-concept and Text Categorization Technology

• Main Author: 劉智凱
• Other Authors: 翁頌舜
• Format: Others
• Language: zh-TW
• Published: 2003
• Online Access: http://ndltd.ncl.edu.tw/handle/06605847480716191389

碩士 === 輔仁大學 === 資訊管理學系 === 91 === 在文件探勘的領域中，文件分類技術的應用範圍可說是非常的廣泛，其主要的應用範圍包括了文件過濾，字詞辨義，網頁分類等。透過文件分類的機制，可以將文件歸類至事先定義好的類別中以減少人們花費在搜尋文件上的時間成本。 本研究則將文件分類技術應用於郵件的回覆範本推薦上，藉此減輕客服人員在回覆郵件時的負擔，藉由範本的推薦使得客服人員在其限定的範本個數中尋找所需的解答範本，進而減少在找尋相關解答時的過多資訊與時間的浪費。 然而目前文件分類機制仍以單一概念為主，本研究希望藉由多重概念的觀念結合概念與類別間的相關性，達成將文件分類的功能。透過不同概念與類別間的組合，能夠動態對具有同一概念的郵件，推薦不同的回覆範本，藉此能夠確實表現出郵件間的差異性，有效的提升推薦範本時的準確性。 此外，本研究亦嘗試對詢問兩個以上問題的郵件推薦其適當的回覆範本，藉由實驗驗證，本研究所提出之方法對多問題郵件在範本的推薦上具一定的效果，故採用多重概念來呈現文件主題的方式，確實較利用相似文件的向量來呈現文件主題更能清楚的擷取文件所要傳遞的資訊。 ABSTRACT Keywords: Text Mining, Text Categorization, Multiple Concepts, Template Recommendation. In the domain of text mining, the application of text categorization is very popular. The scopes of this application include text filtering, word identification, web page categorization, and so on. The techniques can classify the documents into specified classes and help us to find documents fast. This research applies the techniques of text categorization to recommend e-mail templates of replying customers’ e-mails, for the purpose of reducing the burdens of customer service representatives who are responsible to reply customers’ e-mails. However, the idea of single concept is still the main focus on the study of text categorization. Our aim is to combine the idea of multiple concepts and the relationships between concepts and classes to achieve the enhancement of text categorization. According to different combinations of concepts and classes, we can recommend different replying patterns to e-mails that have the same concepts dynamically. The result shows that our approach can differentiate the e-mails and improve the precision to reply e-mails effectively. Besides, we also try to recommend appropriate templates for mails that contain more than two questions. We have proved that our method has a good result in recommending patterns and use multiple concepts to represent mail topics more effectively.