基於職業技能和教育視訊之學習內容生成與總結方法;Learning Content Generation and Summarization based on Industrial Skills and Educational Videos

NCUIR > College of Electrical Engineering & Computer Science > Graduate Institute of Computer Science and Information Engineering > Electronic Thesis & Dissertation > Item 987654321/86434

Please use this identifier to cite or link to this item: http://ir.lib.ncu.edu.tw/handle/987654321/86434

Title:	基於職業技能和教育視訊之學習內容生成與總結方法;Learning Content Generation and Summarization based on Industrial Skills and Educational Videos
Authors:	佳樂恩;Chootong, Chalothon
Contributors:	資訊工程學系
Keywords:	學習內容推薦;知識管理;自動生成內容;計算機科學課程2013;教育視頻摘要;提取總結;字幕總結;集成CNN-LSTM;多注意力機制;教育視頻摘要;Learning Content Recommendation;Knowledge Management;Auto Generated Contents;Computer Science Curriculum 2013;Lecture Video Summarization;Extractive Summarization;Subtitle Summarization;Integration CNN-LSTM;Multiple Attention Mechanism;Educational Video
Date:	2021-06-09
Issue Date:	2021-12-07 12:50:04 (UTC+8)
Publisher:	國立中央大學
Abstract:	資訊與通信科技（ICT）領域產業的需要的知識技能大量增加。由於資訊產業工作的內容廣泛，若要教導學生全面的技能使之可以用於各種不同產業的工作是不太可能的。學生在大學所學的課程內容與產業需求也就存在著落差的議題，因此，我們提出此學習內容推薦系統（LCRec），讓學生可以依據工作需要的技能去媒合課程內容。本研究的第一個目的是提供一個整合IT相關工作所需技能、維基百科、2013年計算機科學課程單元（CS2013）的技能手冊。我們透過公開的求職網站進行產業需求的調查，調查結果也可以讓大學有效的暸解產業所需的技能，並進一步去強化相關技能課程的學習內容。我們透過專家學者及學生對我們提出的學習內容推薦系統（LCRec）實用性進行實驗並分析回饋結果。研究結果表明，本系統可以有效的減少大學與產業間的學用落差（即學習內容不足）現今學習的內容的來源不只僅是書籍，還包含視頻、部落格、網頁等。值得注意的是教育視頻已是人們獲取新知的重要媒體。然而許多視頻都缺少內容的描述。本研究的第二目標為，提出一自動字幕摘要機制。基於多重注意力機制，透過卷積神經網絡 (CNN) 和雙向長短期記憶 (Bi-LSTM) 神經網絡之混和模型，推測句子的關鍵信息。在實驗階段中，文檔中的每個字幕句子，都被分配到一個顯著分數，並依據句子之特徵及其對應之分數作訓練學習，進而產生視頻摘要。此外，本研究於實驗階段，蒐集DUC2002 和 CNN/Daily Mail文本文檔數據集並進行訓練，測試我們模型的性能。由 ROUGE 度量來評估生成的摘要，取得 95% 的置信區間之實驗結果可以明顯發現，我們的模型在 ROUGE-1、ROUGE-2 和 ROUGE-L 分數上優於其他先進的模型;Knowledge skills in Information Communication Technology (ICT) industry always emerge. With the wide variety of jobs available, it is unlikely to educate students who have all skills to match every job requirement. This issue strongly indicates gaps between what is taught in the university and what the industry needs. Therefore, we propose the Learning Content Recommender (LCRec) for students to find appropriate learning contents based on required job-skills. The first purpose of this research is to provide skill books that are hybridization on IT job-skills, Wikipedia, and Knowledge Units from the Computer Science Curriculum 2013 (CS2013). Skills from publicly available job searching websites are used to investigate what the industry needs. Moreover, it is also useful for academics to look at the skills needed in industries and consider enhancing the curriculum with new skills. We carried experiments and analyzed the feedback among professionals, academics, and students to test the usefulness of LCRec. The study result demonstrated that it is possible to bridge the gap (what learning contents are lacking) between the academia and the industry. The learning content is not only books but also includes videos, blogs, webpage, etc. Notably, the educational video is an essential material for people to update new knowledge. However, many videos lack the description that might consume the user time to get a suitable video and gain the core content of the video. This is an inspiration to study an automatic subtitle summarization. For the second goal, we introduce a novel multiple attention mechanism for subtitle summarization. Both advantages from Convolutional Neural Networks (CNNs) and Bidirectional Long Short-Term Memory (Bi-LSTM) Networks are utilized to capture the critical information of the sentence. Each sentence in the subtitle document is assigned a salient score and then video summaries are produced based on sentence feature and its score. The experiments are conducted on both subtitle documents from educational videos and text documents. Besides, we experiment on two well-known text document datasets, DUC2002, and CNN/Daily Mail, to test the performance of our model. We utilize ROUGE measures for evaluating the generated summaries at 95% confidence intervals. The experimental results demonstrated that our model outperforms the baseline and state-of-the-art models on the ROUGE-1, ROUGE- 2, and ROUGE-L scores.
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	75	View/Open

社群 sharing

Loading...