運用卷積神經網路偵測網站頁面異常研究;Detecting Abnormal Website Pages by Convolutional Neural Networks

NCUIR > School of Management at National Central University > Graduate Institute of Information Management > Electronic Thesis & Dissertation > Item 987654321/84002

Please use this identifier to cite or link to this item: http://ir.lib.ncu.edu.tw/handle/987654321/84002

Title:	運用卷積神經網路偵測網站頁面異常研究;Detecting Abnormal Website Pages by Convolutional Neural Networks
Authors:	賴之康;Lai, Chih-Kang
Contributors:	資訊管理學系
Keywords:	網頁;跑版;影像辨識;深度學習;卷積神經網路;web;web page;Abnormal Website;Image recognition;deep learning;CNN
Date:	2020-07-10
Issue Date:	2020-09-02 17:53:15 (UTC+8)
Publisher:	國立中央大學
Abstract:	現代的人，使用電腦或行動裝置上網已經是每天習慣要做的事，瀏覽的網站從一般的內容型網站、社群網站、影音媒體網站到電子商務型網站都有，應該有人碰過，進到某個網站就發現頁面錯誤或是呈現的網頁內容是壞掉的，可能少了一張圖片、圖片與文字對不起來或是某個區塊跑到了不該出現的地方，這樣的狀況在業界稱之為跑版；相信建置網站的開發團隊都極不願意把這些跑版的資訊呈現在使用者的眼前，這樣不僅可能會流失網站的流量，最重要的是讓自己網站的品質受到了質疑與傷害；本研究主要在探討使用圖片/影像辯識的方法，對網站頁面轉成的圖片進行辨識是否有跑版的問題發生；實驗中使用深度學習在影像辨識領域表現得最好的卷積神經網路演算法，搭配圖片數量、圖片尺寸、訓練回合數、卷積層數等變因進行訓練，根據本研究實驗得到的結果顯示，若各變因有適當的調整，則所獲得的準確率及混淆矩陣分類正確性都會獲得良好的改善。;Modern people, using computers or mobile devices to surf the Internet is a habit to do every day. The websites browsed are from general content sites, community sites, video sites to e-commerce sites. In some cases, when we visit a website, some webpage layout is incorrect or the content of the presented webpage is broken. There may be a missing picture, image and non-matched text, or a content block has moved to a place where it should not appear. This situation is called "broken layout" within the industry. It is true that the development team that built the website is very reluctant to show the "broken layout" to end users, so that not only may the website traffic be lost, but also the degradation for the quality of the website. Users would have doubt about the quality and hurt brand. This research is mainly to explore the method of using image identification to identify whether the image converted from the website page has a "broken layout" problem; deep learning is used in the experiment since it performs well in the field of image identification. The neural network algorithm is trained with different factors such as the number of training pictures, the size of the pictures, the number of training iterations, and the number of convolutional layers. According to the results of this research, if the various factors are adjusted appropriately, the accuracy rate obtained and the confusion matrix classification accuracy will be improved.
Appears in Collections:	[Graduate Institute of Information Management] Electronic Thesis & Dissertation

Files in This Item:

File	Description	Size	Format
index.html		0Kb	HTML	120	View/Open

社群 sharing

Loading...