網頁層級資料擷取系統;Page-level Information Extraction System

NCUIR > College of Electrical Engineering & Computer Science > Software Engineer > Electronic Thesis & Dissertation > Item 987654321/68576

Please use this identifier to cite or link to this item: http://ir.lib.ncu.edu.tw/handle/987654321/68576

Title:	網頁層級資料擷取系統;Page-level Information Extraction System
Authors:	丁中立;Ding,Jhong-li
Contributors:	軟體工程研究所
Keywords:	資料擷取;地標;使用者介面
Date:	2015-07-29
Issue Date:	2015-09-23 12:51:42 (UTC+8)
Publisher:	國立中央大學
Abstract:	在網頁資訊擷取(Web Data Extraction)的領域中，如何自動的從各種不同架構的網頁中擷取資料的相關議題至今已被探討研究十年，然而由於現今網頁的內容多樣與架構的複雜，現有的方法均有其限制之處，再加上大量網頁擷取的需求，使得網頁資訊擷取的研究仍面臨相當大的挑戰。網頁資料擷取系統主要分成記錄層級(Record Level)和頁面層級(Page Level)兩大類別，雖然頁面層級相較於記錄層級能夠得到更完整的網頁資訊，但由於問題的複雜及實作的困難，使得現今提出的系統中，其擷取的效能與效率都有改進的空間，此外現存系統皆需要使用者具有資訊背景，沒有提供簡單友善的圖形介面(GUI)。在本篇論文當中，我們提出了一套頁面層級資訊擷取系統，M.-C. Chen及T.-S. Chen所提出的頁面層級系統的架構為基底，提供一個簡單友善的圖形介面，讓使用者，可以用此系統，快速擷取出所需要的網頁資訊。並且再往上對其訓練的流程做改良，以提升系統的擷取效能；在本論文的實驗中顯示，對於訓練的流程上的改良結果，不但不影響原本在表列網頁(List Page)就很好的部份，且在詳細網頁(Detail Page)中，準確率(Precision)提升了33.08%、召回率(Recall)提升32.4%，在整體效能比較中，改善後的系統得到了最高的召回率。在精確度(Accuracy)部份，實驗顯示改良後的系統光是預設的模組參數值，在整體精確度就比TEX還要高出許多；若是再以人工調整模組參數，整體精確率可再向上提升至98.8%，整體精確率比TEX還要高27%。 ;The problem of web data extraction has been studied more than ten years. Because of the structural complexity and diversity in web pages, existing researches are limited to record-level data extraction. Beside, demand of extracting data from large amount of web pages make it a challenging task for researchers. Although the web data extracted by page-level approach is more complete than record-level approach, very few researches focus on this task because of the difficulties and complexities in the problem. On the other hands, existing web data extraction systems need IT background users, because these systems have not provide friendly GUI for users. In this pager, we provide a web data extraction systems based on M.-C. Chen and T.-S. Chen. We provide a friendly GUI for users to improve the training procedure of the schema induction process. The experimental results show that the performance on list page websites remain high and the performance on detail pages are increased precision 33.08% and recall 32.4%. In addition, improved system get highest recall than other systems. For accuracy, our system is higher than TEX with default threshold. If we adjust the threshold of models, we can improve the overall accuracy form 94.5% to 98.8%; Overall accuracy is 27% higher than TEX.
Appears in Collections:	[Software Engineer] Electronic Thesis & Dissertation

Files in This Item:

File	Description	Size	Format
index.html		0Kb	HTML	794	View/Open

社群 sharing

Loading...