應用於中小學問題式學習結合卷積神經網路與支持向量機器之問題分類器;A Self-relevant CNN-SVM Model for Problem Classification in K-12 Question-driven Learning

NCUIR > College of Electrical Engineering & Computer Science > Graduate Institute of Computer Science and Information Engineering > Electronic Thesis & Dissertation > Item 987654321/72027

Please use this identifier to cite or link to this item: http://ir.lib.ncu.edu.tw/handle/987654321/72027

Title:	應用於中小學問題式學習結合卷積神經網路與支持向量機器之問題分類器;A Self-relevant CNN-SVM Model for Problem Classification in K-12 Question-driven Learning
Authors:	何紹儀;Ho,Shao-Yi
Contributors:	資訊工程學系
Keywords:	分類器;卷積神經網路;支持向量機器;Word2Vec;問題式學習;Classification;Convolutional Neural Network;Support Vector Machine;Word2Vec;Question-driven Learning
Date:	2016-07-18
Issue Date:	2016-10-13 14:22:26 (UTC+8)
Publisher:	國立中央大學
Abstract:	隨著科技的發展與進步，我們所能接觸到的資訊已不局限在生活周遭，以學習為例，傳統的教育方式，是由老師在教授並規劃學習的內容與進度，但網際網路的發達，藉由電腦與手機，我們能接收更多元的教材。在問題式學習中，希望能透過問題的方式來輔助學生學習以及讓他們更了解一個問題背後的知識含意。如此多元的資料來源，就需要一個好的管理，以知識為主要目標，將問題依照不同的內容來做分類，也可避免同一個知識點因為不同的表達方式而有不同的定義範圍。本篇論文提出了一個針對中小學學習教材的分類系統。現代教材中的題目如果單單只用文字表示，有時無法清楚的表達正確的意思。所以大部分的題目會加入影像或圖片來輔助，我們的系統也針對這樣的情況，對文字與圖片分別設計了處理的模組。在分類器系統上，設計了結合卷積神經網路(Convolutional Neural Network)與支持向量機器(Support Vector Machine)架構的一個分類器模組，並使用Word2Vec作為卷積神經網路的嵌入層之權值。藉由文字與圖片資料的整合處理，以及結合兩個分類器模組的系統設計，使得分類效果能夠優於原本只使用文字做分類和只使用單一分類器之設計。 ;With the development and progress of science and technology, the information we can get is no longer limited around us. Take learning as an example, arranging the learning content and progress is done by teachers in the traditional learning method. But after the appearance of Internet, we can get a different kind of learning materials online. In Question-driven learning, students clarify and validate what they learn through answering questions. Such big amount of questions needs good management. A well-performed management could avoid the scene that learning materials with same knowledge set are defined in different section due to ambiguous expressions. In this paper, we proposed a hybrid classification model using Convolutional Neural Network with Support Vector Machine that focuses on K-12 learning materials. We utilize Word2Vec as the initial weights for the embedding layer of CNN. In response to the current question not only contains text but also appear with image, we introduced a multi-modal preprocessing approach that first handles text and image separately, then convert them into single modality for classification. The experiment shows that the multi-modal preprocessing and the hybrid model can outperform single-modal preprocessing and individual classification model.
Appears in Collections:	[Graduate Institute of Computer Science and Information Engineering] Electronic Thesis & Dissertation

Files in This Item:

File	Description	Size	Format
index.html		0Kb	HTML	485	View/Open

社群 sharing

Loading...