Automatically Discovering Relevant Images From Web Pages

Erdinc Uzun,Erkan Ozhan,Tarik Yerlikaya,Halil Nusret Bulus,Hayri Volkan Agun

doi:10.1109/access.2020.3039044

Erdinc Uzun, Erkan Ozhan + Show 3 more

Open Access

https://doi.org/10.1109/access.2020.3039044

Copy DOI

Journal: IEEE Access	Publication Date: Jan 1, 2020
Citations: 39	License type: CC BY 4.0

Affiliation: Namık Kemal University, Bursa Technical University

Abstract

Web pages contain irrelevant images along with relevant images. The classification of these images is an error-prone process due to the number of design variations of web pages. Using multiple web pages provides additional features that improve the performance of relevant image extraction. Traditional studies use the features extracted from a single web page. However, in this study, we enhance the performance of relevant image extraction by employing the features extracted from different web pages consisting of standard news, galleries, video pages, and link pages. The dataset obtained from these web pages contains 100 different web pages for each 200 online news websites from 58 different countries. For discovering relevant images, the most straightforward approach extracts the largest image on the web page. This approach achieves a 0.451 F-Measure score as a baseline. Then, we apply several machine learning methods using features in this dataset to find the most suitable machine learning method. The best f-Measure score is 0.822 using the AdaBoost classifier. Some of these features have been utilized in previous web data extraction studies. To the best of our knowledge, 15 new features are proposed for the first time in this study for discovering the relevant images. We compare the performance of the AdaBoost classifier on different feature sets. The proposed features improve the f-Measure by 35 percent. Besides, using only the cache feature, which is the most prominent feature, corresponds to 7 percent of this improvement.

Highlights

We developed a web crawler1 to download web pages and their images for creating a new dataset2
That is, having irrelevant features in your dataset can decrease the performance of the models constructed from machine learning methods
After all images are processed for a web page, new features are computed for those images

Summary

INTRODUCTION

T HE internet provides us with a lot of valuable content among HTML tags. Nowadays, extracting this content automatically is one of the challenging tasks in information retrieval. For determining the most representative image on a web page, Vyas and Frasincar [9] construct their prediction model on the machine learning method, Support Vector Machines (SVM). These studies are not contributive in terms of the comparison of various classification methods. We report the best evaluation score along with the comparison of several different machine learning methods in discovering the relevant images. The image datasets constructed for relevant image extraction is imbalanced This is known as an imbalanced dataset problem that negatively affects the prediction model of machine learning methods. That is, having irrelevant features in your dataset can decrease the performance of the models constructed from machine learning methods.

RELATED STUDIES

CRAWLER AND FEATURES

Calculate Features for a given image

LAST FEATURES

EXPERIMENTS

Findings

CONCLUSIONS