透過您的圖書館登入
IP:3.149.229.253
  • 學位論文

學術論文摘要的自動文步分析

Automatically Identify Moves in Academic Abstracts

指導教授 : 張俊盛
若您是本文的作者,可授權文章由華藝線上圖書館中協助推廣。

摘要


在本論文中,我們利用督導式機器學習方法,經由自動抽取論文摘要中重要的特徵值,自動建立文步標示與特徵值間的關係並以此訓練機器學習模型。以此機器學習模型可以自動標示論文摘要中的文步結構。 在訓練階段,我們利用少量人工標記文步的論文摘要,從中取出重要的特徵值,再訓練機器學習模型學習特徵值與文步之間的關係,以進行對論文摘要的自動文步分析。在執行階段,應用訓練好的機器學習模型,我們分別針對電腦科學領域與應用語言學領域標示研究者所撰寫的論文摘要文步,我們利用人工評估的方法,在兩個不同領域下,皆可達到平均超過80%的正確率,顯示本方法可以成功標記論文摘要文步。

並列摘要


This paper presents a novel method for automatically identifying the move structure in academic abstracts to assist non-native speaker of English in academic writing. In our approach, we use a small set of manually tagged abstracts as training corpus and analyze the significant features. Maximum Entropy model (ME) is employed to classify the move structure in the given abstracts. It involves automatically learning of the syntactic features, and automatically building a statistical model. The proposed method outperforms the previous research with a significantly higher accuracy. Our methodology clearly shows that the ME could suitably model the abstract structure, and implies that a more flexible move tagger can be easily applied to different research domains using a small set of manually tagged abstracts.

參考文獻


introductions in two disciplines. English for specific purposes, 24(2), 141-156.
ANSI. (1979). American national standard for writing abstracts. Z39.14-1979,
American National Standards Institute (ANSI).
Anthony, L. & Lashkia, G. V. (2003). Mover: A machine learning tool to assist in the
reading and writing of technical papers. IEEE Trans. Prof. Commun., 46, pp.

延伸閱讀