電子郵件是現代人最重要的通訊工具之一，不論是在工作上或是一般生活中，每天都會收到許多的電子郵件。而電子郵件中往往有許多重要的資訊，譬如會議或行程邀約郵件就會有事件時間的重要資訊，這些資訊若不經過人工的判別，並且手動將這些行程標註於行事曆中，則很可能就會讓此類重要資訊淹沒於大量郵件當中，而導致錯過重要的行程。面對此種問題，人們需要一套自動化的解決方案，但是郵件內容為非結構化文件，不易辨識是否為行程邀約，並且其中的時間，多是口語性的表達，亦不易辨識及擷取。因此本研究希望建構一套系統，能夠辨識行程邀約郵件，再將這些行程邀約郵件中的時間表達字串擷取出來，做為日後提醒之依據。本系統分為兩個部份，第一部份是擷取郵件的特徵，藉由支持向量機分纇器，訓練出分類郵件的模型，來辨識行程邀約郵件。第二部份是將這些郵件中的時間資訊，採用條件隨機場域，訓練出標記時間表達字串的模型來萃取時間關鍵字，最後系統再透過Google Task API自動地將萃取出的行程加入於Google Task中。此機制可以減輕使用者人工判別的負擔，亦減少了錯失行程的機會。實驗結果顯示，本系統所提出之方法在邀約郵件的辨識上可達94.8的F-measure，在時間擷取上也可達到95.7的F-measure。Nowadays, E-mail reader is one of the most important communication tools. Many people receive a lot amount of e-mails in business or in daily life. Invitation e-mails often contain important information that need to be tracked for some time. Such messages might be forgotten easily if people do not handle it immediately and mark on their calendar right away. To deal with this issue, people need an automatic solution which can recognize invitation e-mails and the time expressions for later reminders. The challenge here is information extraction from non-structure free text.This research proposed a system that would be able to recognize invitation e-mails and extract the time expressions via machine learning. This system is composed of two parts. The first part is utilizing the Support Vector Machine classifier to build a model that can predict the class of a new e-mail. The second part is utilizing Conditional Random Fields to build a model that can extract time expression from an e-mail. Finally, we can extract the time expression and append it in Google Task Service by using Google Task API. This mechanism can reduce the effort of reading e-mails and decrease the opportunity of missing events. The proposed methods achieve a 94.8 and 95.7 F-score for recognizing invitation e-mail and time expressions, respectively.
