International Journal of Computer Science Research and Application 2012, Vol. 02, Issue. 03, , pp. 02-12 ISSN 2012-9564 (Print) ISSN 2012-9572 (Online) © Author Names. Authors retain all rights. IJCSRA has been granted the right to publish and share, Creative Commons 3.0

INTERNATIONAL JOURNAL OF COMPUTER SCIENCE RESEARCH AND APPLICATION www.ijcsra.org

Automated Online News Content Extraction Ojokoh, Bolanle Adefowoke [email protected] Author Correspondence: Department of Computer Science, Federal University of Technology, P.M.B. 704, Akure, NIGERIA

Abstract With the growth of the Internet and related tools, there has been an exponential growth of online resources. This tremendous growth has paradoxically made the task of finding, extracting and aggregating relevant information difficult. These days, finding and browsing news is one of the most important internet activities. In this paper, a hybrid method for online news article contents extraction is presented. The method combines RSS feeds and HTML Document Object Model (DOM) tree extraction. This approach is simple and effective at solving the problems associated with heterogeneous news layout and changing content found in many existing methods. The experimental results on some selected news sites show that the approach can extract news article contents automatically, effectively and consistently. The proposed method can also be adopted for other news sites.

Keywords: Online news; Information extraction; RSS feeds; Title; HTML; Document Object Model; Search Engine

1. Introduction The Web is the largest data repository ever available in the history of human kind. It has become the richest source of information. Although there is a tremendous amount of information available, they are not always in the forms that support end-users’ needs. There is a growing trend of enabling users to view diverse sources of data in an integrated manner. Major efforts have been made in order to provide efficient access to relevant information within this huge repository of data. Some of these efforts have been tailored towards document metadata extraction (Thoma, Mao and Misra, 2005) and reference extraction (Ojokoh et al., 2011; Seymore, McCallum and Rosenfeld, 2009) among others. Information extraction has become an important technology to help users locate desired information on the Web. Data extracted from Web sites can serve as the springboard for a variety of tasks, including information retrieval (e.g. business intelligence), event monitoring (news and stock market), and electronic commerce (shopping comparison). Extracting structured data from Web sites is not a trivial task. Most of the information on the Web today is in the form of Hypertext Mark-up Language (HTML) documents which are viewed by humans with a browser. HTML documents are sometimes written by hand, sometimes with the aid of HTML tools. Given that the format of HTML documents is designed for presentation purposes, not automated extraction, and the fact that most of the HTML content on the Web is ill-formed (“broken”), extracting data from such documents can be compared to the task of extracting structure from unstructured documents (Myllymaki, 2001). The Web is rapidly moving towards a platform for mass collaboration in content production and consumption, and an increasing number of people are turning to online source for their daily news. Traditional newspapers have developed significant Web presence. Fresh contents on a variety of topics, people, and places

3

International Journal of Computer Science Research and Application, 2(3): 02-12

are being created and made available on the Web at a breath taking speed. With thousands of new websites providing daily news services (e.g. Google News and Yahoo! News) in today’s Web, coupled with the fact that news is among the most popular interests for Web surfers, it is critical to provide a tool that can automatically extract online news information for users (Chen et al., 2009). The concept of online news extraction means generating structured information from unstructured/semi-structured data which is retrieved from the web. News recognition and news extraction approaches deal with two problems: the identification of news articles pages within a collection of heterogeneous web documents where there are many not desired content (like section pages, headlines, advertisements) and given a document identified like a news page, the extraction of the fields of the article, that is the title, the body (Parapar and Barreiro, 2007). Most of previous news extraction approaches use manually or automatically constructed wrappers to extract news information. Such approaches assume that news information is wrapped by recurring physical or virtual patterns across news pages. One of such approach is Tree Edit Distance (TED) (Reis et al., 2004) which generates wrappers based on the consistency of HTML Document object model (DOM) trees. Another approach is Visual Wrapper (VW), which learns wrappers based on recurring visual patterns (Zheng, Song, and Wen, 2007); they have inappropriate assumptions which may not always hold due to the heterogeneous and dynamic characteristics of the online news information. For example, TED assumes that templates exist and can be extracted based on DOM trees, thus the generated wrapper can only work properly for pages that share a specific template that recurs in DOM trees of Web pages. TED is template-dependent and requires that multiple pages with the same templates exist in the news corpus to be extracted. VW is based on the assumptions of some visual features like font attributes, presence of contiguous text paragraphs which may not always be true due to the diversity of web authoring technique and news content property. Many exceptions exist and result in the ineffectiveness of the approach. VW requires a training stage to derive wrappers. The training data need be labelled manually. Extraction results may not be satisfactory when the training set is too small. These traditional methods have a lot of challenges to combat with. The news sites comprise of different kinds of Web pages. Besides the news pages, there are many non-news pages, such as blog, shopping, weather, advertisement, yellow pages and even same pages with different URLs. Furthermore, these news pages are spread in the different sections of news sites. The news sites are crawled to find as many news pages as possible, but actually, it is difficult to recognize and acquire all the news pages quickly from a large number of Web pages. The different news sites use different news page layouts, and each news site uses more than one layout. Moreover, the news sites update the layout of news pages irregularly. If the news sites update the layout of news pages, the corresponding analysis has to be done again. It is therefore not an easy task to extract news article contents from news sites efficiently and quickly over a long period of time using traditional methods. In this paper, an automatic Web news extraction method that is capable of extracting news article contents from news sites over a long period of time is proposed. It employs Really Simple Syndication (RSS) feeds. RSS is a family of Web feed formats used to publish frequently updated content such as news headlines. The latest news pages from news RSS feeds can be collected conveniently as soon as they are published. The extraction method is independent of Web page layout and does not need to analyze the news sites before extraction. A similarity algorithm is also used to calculate the relevance between the news title and each contiguous sentence in order to detect the news paragraphs from the full text of the news page. It also makes provision for extraction of relevant information based on topic search. This approach is effective at solving the problem of heterogeneity and temporariness associated with other methods. In addition, the method proposed in this paper is not as complex as existing methods and does not need any maintenance during the long period of extraction. The remainder of this paper is organized as follows: Section 2 presents some related works. Section 3 gives a detailed description of the proposed approach for online news content extraction. The implementation and experimental results are presented in Section 4, while Section 5 gives the conclusion and directions for future research.

2. Review of Related Studies News is the information about recent and important events. It is the communication of selected information on current events which is presented by print, broadcast, internet, or word of mouth to a third party or mass audience. Several approaches have been proposed to solving a variety of problems in the news domain. Some of these as described by Sayyadi, Salehi, and Abolhassani (2006) include the work of news collection or extraction, news retrieval (Corso, Gulli, and Romani, 2005), categorization of news search results, news

4

International Journal of Computer Science Research and Application, 2(3): 02-12

summarization and automatic event detection (Naughton, Kushmerick, and Carthy, 2006). Online News Extraction has become a prominent research problem these days, as a lot of methods have been proposed towards addressing this problem. Some rule-based approaches have been adopted by a number of researchers. Parapar and Barreiro (2007) proposed a set of heuristics related to document styling to identify and extract information article from news, or reject a not-news web page. The implementation of these heuristics results in a linear complexity algorithm on the web page length. The linear complexity algorithm follows the content of the HTML documents looks for paragraphs that match with the criteria and joins them to build the news body. It also requires setting the values of some parameters like paragraph minimum size, news body minimum size, inter-paragraph maximum distance and hyperlink density. The method is template-dependent. Guo et al. (2010) proposed a method that uses a DOM tree to represent the Web news pages. The method finds a snippet-code by which a part of the content of news is wrapped firstly, and then backtracks from the snippet-node until a summary-node is found, and the entire content of news is wrapped by the summary-node. Dong et al. (2008) gave a generic web news article contents extraction approach based on a set of predefined tags. The method does not need to find any hidden template for each page, but instead uses a few heuristics based on some identified observations about the DOM trees from articles on news web pages. Some of them include the fact that news articles, (including text, date and title) are generally located under a separate node; they are comprised of a number of paragraphs, located close to each other and often with other, unrelated material, between them. Format wise, they contain a lot of text and few links. The experiment of this method is based on the assumption that the news pages from a news site use the same layout. But, actually, there are many different layouts used in a news site. The tree edit distance metric is a common similarity measure for rooted ordered trees. Since the structure of a Web page can be described by a tree (e.g. a Document Object Model tree), this method has found its use in solving the news extraction problem. Reis et al. (2004) utilized the concept of tree edit distance to evaluate the structural similarities between pages. The work recognized and explored some common characteristics that are usually present in news portals and tried to correctly extract the news, disregarding the other pages. It also relied on the basic assumption that the news site content can be divided in groups that share common format and layout characteristics. Tree edit distance might be costly to adopt in a highly heterogeneous environment. Zheng, Song, and Wen (2007) solved the problem of news extraction by adopting a template-independent approach. They viewed a news page as a visual block tree, derived a composite visual feature set by extracting a series of visual features and used supervised machine learning with manually labelled training data to generate a wrapper for extraction. Another, quite similar approach which also focuses on how a human is believed to go about when finding news is given by Chen and Lui (2008) in ‘Perception-oriented online news extraction’. Properties, quite similar to the ones mentioned by Zheng, Song, and Wen (2007) are given to the areas of a page which contain the actual news content, and which humans use to identify news content. Ziegler and Skubacz (2007) suggest a system which identifies text blocks in a document and finds threshold values for different features, (properties) of the blocks to determine if a block is a part of the article or not. The actual values of the thresholds are calculated using a stochastic non-linear optimization method called Particle Swarm Optimization using a large number of pages. These threshold values are then used when extracting other pages. Extraction results may not be satisfactory when training set is too small. Second, even with these prerequisites satisfied, the extraction results may still be unstable and domain/site dependent. Lindholm (2010) proposed a generic algorithm based on Vector Space Model (VSM) that organizes documents according to the terms that they contain for content extraction from online news sources. He particularly contributes to an investigation of how adding syntactical information to VSM affects search results. With the method, no special wrappers need to be generated, the only site specific information that must be stored about each site are filters and multi-page tags if such are applied. However, the precision results are quite low and the method may not perform well outside common newspaper articles without too much formatting and in new sites with too much additional information such as comments blog pages. In essence, the extraction results may be affected by the time of publication. Zhang and Lin (2010) proposed automatic template-dependent Web news content extraction approach based on similar pages. In the work, two similar pages were chosen as training samples and represented as two HTML DOM trees. After these, the maximum matching tree between the DOM trees was created using simple tree matching and backtracking algorithms. In an attempt similar to Guo et al. (2010), they eliminated the noise nodes to generate an extraction template by analyzing the characteristics of nodes in the maximum matching tree. Finally, they built a template-dependent wrapper for target news pages whose structures are similar to the samples. RSS based methods have also found some use by a number of research works. For instance, Adam, Bouras, and Poulopoulos (2010) proposed an incremental crawling mechanism that supports personalized RSS feeds

5

International Journal of Computer Science Research and Application, 2(3): 02-12

using a learning approach for offering collections of news articles in real time to internet users. Training could be expensive and ineffective when the data set is few. The two major steps taken in the work are training and crawling, but their data sets consist of RSS feeds. So, their approach is quite different from the one employed in this paper. Han, Tomoya, and Takehiro (2009) proposed a related RSS based approach to extraction of Web News Article Contents Extraction. To address these problems, an approach to realize the automatic news article contents extraction based on RSS feeds is proposed. News pages can be collected from the news RSS feeds easily because more and more news sites distribute the latest news by RSS feeds. RSS feeds extracts automatically. It has a high extraction precision over a long period of time. In combination with RSS feeds, some other algorithms were proposed for full-text analysis, news title detection and news content extraction. These methods are different from the ones adopted in this work. This work combines HTML DOM tree with RSS feeds. The method proposed in this paper is not as complex and does not still need any maintenance during the long period of extraction.

Figure 1: Architecture of the system

3. Online News Content Extraction Figure 1 presents the architecture of the system. News pages are a collection of web pages obtained from different online news websites. The RSS feeds are obtained from the RSS feed URL from each web news page. The several steps followed in the adopted approach are described in the following sub sections.

3.1 Description of RSS Feeds RSS (most commonly expanded as Really Simple Syndication) is a family of web feed formats used to publish frequently updated works—such as blog entries, news headlines, audio, and video—in a standardized format. An RSS document (which is called a "feed", "web feed", or "channel") includes full or summarized text, plus metadata such as publishing dates and authorship. Web feeds benefit publishers by letting them syndicate

6

International Journal of Computer Science Research and Application, 2(3): 02-12

content automatically and readers who want to subscribe to timely updates from favoured websites or to aggregate feeds from many sites into one place (Han, Tomoya, and Takehiro, 2009). RSS was designed to show selected data. Without RSS, users will have to check the site daily for new updates. This may be too time-consuming for many users. RSS removes spam. Since RSS data is small and fast-loading, it can easily be used with services like cell phones or PDAs. Web-rings with similar information can easily share data on their web sites to make them better and more useful. RSS is a better way to be notified of new and changed content. Notifications of changes to multiple websites are handled easily, and the results are presented well organized. A RSS Document is an XML (Extensible Markup Language) format document. XML is a set of rules for encoding documents in machine-readable form. The design goals of XML emphasize simplicity, generality, and usability over the Internet. Each item in the xml file usually consists of a simple title describing the item along with a more complete description and a link to a web page with the actual information being described. Sometimes this description is the full information one wants to read (such as the content of a weblog post) and sometimes it is just a summary, the publication date and the time of publication.

3.2 Item node detection The RSS feeds are parsed in order to detect and extract the item nodes. An item node contains the news title, link, description, publication date and time. From these links, the HTML document of each news page is extracted from the news site. The real contents are then extracted from the HTML document as described in the following sections.

3.3 Title keywords acquisition The news title is a piece of important information for the recognition of the news contents from the full text of news page. The news title is discovered from the news page, when the item node has been detected. The item node is embedded in an open and a close title tag in the item node. The method adopted in Han et al. is used to obtain the title keywords. When the position of the title in a news page is located, the position of news contents text would be found easily because the contents text is a list of paragraphs closely preceded by the title. Also, for a news article, the contents describe the same topic of news title in detail, and the words constituting the title would occur in the news contents frequently usually. Therefore, the title sentence is split into single words to make a list of keywords and these keywords are used to find out the news article contents from the news page. A keyword list, L is created from the news title using the following steps: 1. Split the news title sentence into a list of words, L using whitespace as the delimiter. 2. Remove all the articles, prepositions and conjunctions from the list of words, L(These include words such as with, or, to, on, of, by and so on). 3. Remove the characters, “’s” or “s” from the words ending with “’s” or “s” in list L.

3.4 HTML Document Analysis HTML Parser is used to analyse the HTML document using DOM (Chen, 2011). Every document may be viewed as a tree structure. Every sentence in the web page can be seen as a leaf node. It is possible for each of these sentences to be the title or a paragraph. At the end of this stage, a fully analyse document is produced. Details about the theoretical background can be found in Guo et al. (2010). The algorithm describing the work of the HTML Parser is described below. Algorithm of the HTML Parser 1.

Begin

2.

Input link

3.

Input description

4.

Call method trimdesc (description)

/* first hundred characters*/

7

International Journal of Computer Science Research and Application, 2(3): 02-12

5.

Load html_doc from link

6.

Storeparagraph = Getpar (loaded html)

7.

N = countparagraph (store paragraph)

8.

M=0

9.

If (M < N) THEN

10.

If (contains (paragraph(M))), trimdesc

11.

Print paragraph (M)

12.

End if

13.

M = M+1

14.

Go to 9

15.

End if Stop

3.5 News Paragraph Recognition Usually, the news contents text is a list of paragraphs immediately below the title. It becomes easier to recognize the paragraphs after the news title in the news page has been found. The method adopted in Han, Tomoya, and Takehiro (2009) was used to solve this problem. In this work, the analysed document is divided into one or more contents ranges and at most one reserve range as follows: If no node is judged as the possible title node, the whole part of the document is a contents range (It has no reserve range). If only one node is judged as the possible title node, the document is divided into one contents range and one reserve range: the following part of the possible title node is a contents range and the preceding part is a reserve range. If two or more nodes are judged as the possible title nodes, the document is divided into some contents ranges and a reserve range: the part between each possible title node (except the last possible title node) and the next possible title node is a contents range, the following part of the last possible title node is a contents range, and the preceding part of the first possible title node is a reserve range. A node whose value is a paragraph is a paragraph node. After the paragraph recognition, the several paragraphs of the entire news contents text are obtained.

3.6 Search Engine The search engine automatically accesses feeds from the websites when the extractor is launched and refreshed, collecting the latest news from the RSS feeds. It searches the collected news using the search keywords. It displays the news title, description, date, and time categorizing by the name of the website from which they are found. It also allows for multiple keyword searches.

4. Implementation and Experiment Results The implementation of the proposed approach and some experimental results on evaluation of the system are presented in this section. For generality, the system was implemented using a large number of news pages from some selected news sites across different continents. These include five sites {The Yahoo News, The CNN, The Brazil News, Soccer News, and African Nigeria News}. The experiments are run over a period of four weeks using randomly collected news articles from those different news sites. An evaluation is carried out manually by comparing the extracted contents with the real

8

International Journal of Computer Science Research and Application, 2(3): 02-12

news articles. The performance of the system is measured using the standard evaluation measures of precision (p), recall (r), F-score (F) and accuracy (a). Table 1 shows the results. The precision (p), recall (r), accuracy (a) and F-measure (F) are evaluated as:

Where, A is the number pages with correctly extracted news article contents (that is, the pages with contents extracted correctly discovered after checking the original news page manually) B is the number of pages with some news contents but is not extracted (that is, the news contents present on the news pages but not extracted by the proposed system) C is the number of pages extracted but with some non-news contents such as advertisements. Figures 1 and 3 show sample extracted pages. Figure 2 show an example of an extracted page categorized under A (that is correctly extracted). This page was extracted from the detailed news page showed in Figure 3. The circled text is the title of the extracted news content. Figure 4 is example of an extracted page categorized under B (that is existing news contents that are not extracted). This page was extracted from the detailed news page showed in Figure 5.

9

International Journal of Computer Science Research and Application, 2(3): 02-12

Figure 2: Sample correctly extracted news page

Figure 3: Detailed news page for figure 1

Figure 4: Sample existing but not extracted news contents

10

International Journal of Computer Science Research and Application, 2(3): 02-12

Figure 4: Detailed news page for figure 2 Table 1. The experimental results of extraction Period

No of Web pages

Precision

F-measure

Recall

Accuracy

Week 1 (July, 2011)

40

92.3%

89.8%

85.7%

80%

Week 2 (August, 2011)

40

97.6%

93.3%

85.1%

83.3%

Week 3 (August, 2011)

40

90%

90%

90%

81.8%

16th August, 2011

40

85.7%

88.9%

92.3%

80%

17th August, 2011

40

92.5%

89.1%

86%

80.4%

Total

200

92.5%

90.2%

87.8%

81.1%

Although, there is a difference in the websites on which the proposed method was tested and many of the existing methods for news extraction, the results obtained can still be compared relatively with some related systems. The two experiments performed by Zheng et al. (2007), who used visual features for online news extraction, showed the effectiveness of the system proposed in this work. V-Wrapper algorithm in the first experiment demonstrated 61.11% average F1, and 88.32% in the second experiment. Both values are less than 90.2% (the F1 result obtained by the system). In addition, the cost of training can be expensive. The experimental results of the DOM implemented by Guo et al. (2010), 41% of the web pages were correctly extracted, 52% were missed, while 7% were wrongly extracted. All these show that thse proposed system is effective at its task. Although the news sites update the layout of news pages irregularly, the proposed news article contents extraction method has good experimental results. For a period of about four weeks, the overall precision, f-

11

International Journal of Computer Science Research and Application, 2(3): 02-12

measure, recall, and accuracy are 92.5%, 90.2%, 87.8% and 81.1% respectively. This proves that the method is consistently effective irrespective of changing news layout. In addition, testing the method using news from different websites across continents also proves the heterogeneity of the method and its application to different news sites. The method proposed in this paper is easier to develop than those applicable in related work.

5. Conclusion In this paper, an hybrid model that combines RSS feeds extraction, HTML DOM parser with a similarity algorithm that identifies the keywords in title nodes and finds the likely similar corresponding paragraph nodes and also a search engine have been developed. This model produces an automatic, effective and efficient method of extracting news article contents from CNN, Soccer news, Brazil news, Africa Nigeria, Yahoo news. The proposed method can also be adopted for other news sites. The experimental evaluation results of several news sites show that this approach works well with a high accuracy rate over a long period of time. As future work, the algorithm will be modified to improve the accuracy rate even further.

References Adam, G., Bouras, C. and Poulopoulos, V.(2010), ‘Efficient extraction of news articles based on RSS crawling’ in: Proceedings of the International Conference on Machine and Web Intelligence pp. 1-7. Chen, J., Wang, J., He, X., Wang, C., Pei, J. and Bu, J. (2009), ‘News article extraction with templateindependent wrapper’ in WWW 2009: Proceedings of the 18th International Conference on World Wide Web, pp. 1085–1086. Chen, J. and Lui, S.C. (2008), ‘Perception-oriented online news extraction’ in JCDL 2008: Proceedings of the 8th ACM/IEEE-CS Joint Conference on Digital Libraries, pp. 363–366. Chen, S.C. (2011) Simple HTML DOM Scraper Script. Available from http://sourceforgenet/projects/simplehtmldom/ (Accessed June 2011). Corso, G.M.D, Gulli, A. and Romani, F. (2005), ‘Ranking a stream of news’ in WWW 2005: Proceedings of the 14th international conference on World Wide Web, pp. 97–106. Dong, Y., Li, Q., Yan, Z. and Ding, Y. (2009), ‘A generic Web news extraction approach’ in: Proceedings of the 2008 IEEE International Conference on Information and Automation, pp. 179–183. Guo, Y., Tang, H., Song, L., Wang, Y. and Ding, G. (2010), ‘ECON: An approach to extract content from Web News Page’ in APWEB 2010: Proceedings of the 12th International Asia-Pacific Web Conference, pp. 314 – 320. Han, H., Tomoya, N. and Takehiro, T. (2009), ‘An Automatic Web News Article Contents Extraction System Based on RSS Feeds’, Journal of Web Engineering; Vol. 8 No. 3, pp. 268–284. Lindholm, S. (2010) Extracting Content from Online News Sites. Unpublished Master’s thesis, UMEA. Sweden. Myllymaki, J. Effective Web Data Extraction with standard XML Technologies 2001. Naughton, M., Kushmerick, N. and Carthy J. (2006) , ‘Clustering sentences for discovering events in news articles’ in ECI 2006: Proceedings of ECI 2006, pp. 535–538. Ojokoh, B., Zhang, M. and Tang, J. (2011), ‘ A Trigram Hidden Markov Model for Metadata Extraction Heterogeneous References’, Information Sciences Vol. 181, pp. 1538-1551.

12

International Journal of Computer Science Research and Application, 2(3): 02-12

Parapar, J. and Barreiro, A. (2007), ‘An effective and efficient Web news extraction technique for an operational news IR system’ in CAEPIA 2007: Proceeding of XIII Conferencia de la Asociacion Espanola para la Inteligencia Artificial II, pp. 319–328. Reis, D.C., Golgher, P.B., Silva, A.S., Laender, A..F and Chen, J. (2004), ‘Automatic Web News Extraction using Tree Edit Distance’ in WWW 2004: Proceedings of the 13th International Conference on World Wide Web, pp. 502 – 511. Sayyadi, H., Salehi, S. and Abolhassani, H. (2006), ‘Survey on News Mining Tasks’ in CIS2E 2006: Proceedings of CIS2E. Seymore, K., McCallum, A. and Rosenfeld, R. (2009). Learning Hidden Markov model structure for information extraction. In AAAI Workshop on Machine Learning for Information Extraction. Thoma, G.R., Mao, S. and Misra, D. 92005). Automated metadata extraction to preserve the digital contents of biomedical collections. In J. J. Villanueva (Ed.), Proceedings of the 5th IASTED International Conference on Visualization, Imaging and Image Processing 2005 (pp. 214-219), Calgary, Canada: ACTA Press. Zhang, C. and Lin, Z. (2010), ‘Automatic web news content extraction based on similar pages’ in: Proceedings of Web Information Systems and Mining, pp. 232-236. Zheng, S., Song, R. and Wen, J-R. (2007), ‘Template-independent news extraction based on visual consistency’ in AAAI 2007: Proceedings of the 22th AAAI Conference on Artificial Intelligence, pp. 1507–1513. Ziegler, C-N. and Skubacz, M. (2007), ‘Content extraction from news pages using particle swarm optimization on linguistic and structural features’ in WI 2007: Proceedings of the IEEE/WIC/ACM International Conference on Web Intelligence, pp. 242–249.

Copyright for articles published in this journal is retained by the authors, with first publication rights granted to the journal. By the appearance in this open access journal, articles are free to use with the required attribution. Users must contact the corresponding authors for any potential use of the article or its content that affects the authors’ copyright.