|摘要: ||好的論文會明確地寫出目的及結論，且兩者之間應具有一致性及完整性，反之則會缺漏造成內容前後不一致，並造成使用者錯誤之引用，使用者搜尋論文主要是透過摘要來評定是否參考，摘要在論文架構中之功能為協助讀者快速理解論文內容，因此好的論文摘要能提升論文搜尋之正確性，摘要內容又以目的及結論為重，故本研究以臺灣碩博論文知識加值系統近六年來中央大學資訊管理系之碩士論文為實驗資料，以TextRank演算法萃取文章之特徵，採用ROUGE-1 做為評量依據評量論文之一致性及完整性，再以多文章之擷取式自動摘要技術TextRank、LexRank、Luhn與潛在語意分析 (Latent Semantic Analysis，LSA)四種方式產生正確性較佳之摘要，與原始摘要做特徵比對，進而評估正確性。
;A good Master’s thesis will clearly stating the purpose and conclusion. Between the purpose and the conclusion, there should be consistency and completeness. On the other hand, missing the above two points, the content will fall short and have contradictions, that will misguide the thesis readers to quote incorrectly. Generally, a reader defines a thesis worth takes references by abstract. The abstract helps the readers to understand the content in a quicker way. Therefore, a good abstract will elevate the correctness of giving a right thesis to meet the reader’s needs. The content of the abstract of a thesis values the purpose and the conclusion the most. This research takes the master’s thesis of the Department of Information Management of National Central University from National Digital Library of Theses and Dissertations System in Taiwan as research data, using TextRank algorithm to extract the features of a thesis, applying ROUGE-1 as evaluation basis to measure the consistency and completeness of the thesis. Furthermore, with the help of the four algorithm of automatic multi-document extraction system TextRand, LexRank, Luhn, and potential semantic meaning analyzation system LSA (Latent Semantic Analysis) to make an abstract with a better correctness. Then, using this automatic summarization from the above technologies to compare with the original abstract to measure the correctness.
The purpose of this research is to comment on the consistency of the paper writing logic, the completeness of the scope, and the correctness of the conclusions, hopefully, after applying the auto- abstract to the original summary, there will be results with better correctness occurring from the thesis searching for the readers. From experiments, the consistency, completeness and correctness of 2013 and 2015 were found better, and 2014 and 2017 were found worse and the professor guidance has a great correlation about 3C structure . The contributions are: (1) Establishing a unsupervised verification system. (2) the consistency, completeness and correctness of the thesis can immediately be assessed pass or not. (3) Provide data for thesis writing to adjust its structure to achieve consistency, completeness and correctness. (4) Professors can automatically check thesis to reduce manual review task. (5) Provide automatic summaries to help improve the accuracy of query thesis.