The Data Records Extraction from Web Pages

International Journal of Trend in Scientific Research and Development (IJTSRD)
Volume 3 Issue 5, August 2019 Available Online: www.ijtsrd.com e-ISSN: 2456 – 6470
@ IJTSRD | Unique Paper ID – IJTSRD28010 | Volume – 3 | Issue – 5 | July - August 2019 Page 2258
The Data Records Extraction from Web Pages
Nwe Nwe Hlaing, Thi Thi Soe Nyunt, Myat Thet Nyo
Faculty of Computer Science, University of Computer Studies, Meiktila, Myanmar
How to cite this paper: Nwe Nwe Hlaing |
Thi Thi Soe Nyunt | Myat Thet Nyo "The
Data Records Extraction fromWebPages"
Published in
International
Journal of Trend in
Scientific Research
and Development
(ijtsrd), ISSN: 2456-
6470, Volume-3 |
Issue-5, August
2019, pp.2258-2262,
https://doi.org/10.31142/ijtsrd28010
Copyright © 2019 by author(s) and
International Journal ofTrend inScientific
Research and Development Journal. This
is an Open Access article distributed
under the terms of
the Creative
Commons Attribution
License (CC BY 4.0)
(http://creativecommons.org/licenses/by
/4.0)
ABSTRACT
No other medium has taken a more meaningful place in our life in sucha short
time than the world-wide largest data network, the World Wide Web.
However, when searching for information in the data network, the user is
constantly exposed to an ever-growing ﬂood of information. This is both a
blessing and a curse at the same time. The explosive growth and popularity of
the world-wide web has resulted in a huge number of information sources on
the Internet. As web sites are getting more complicated, the construction of
web information extraction systems becomes more difficult and time-
consuming. So the scalable automatic Web Information Extraction (WIE) is
also becoming high demand. There are four levels of information extraction
from the World Wide Web such as free-text level, record level, page level and
site level. In this paper, the target extraction task is record level extraction.
KEYWORDS: Information Extraction(IE), Wrapper,DocumentObjectModelDOM
1. INTRODUCTION
The rapid development of World Wide Web has dramatically changed the way
in which information is managed and accessed. The information in Web is
increasing at a striking speed. At present, there are more than 1 billion web
sites and web information has covered all domains of human activities. This
opened the opportunity for users to benefit from the available data. So Web is
being concerned more and more.
To retrieve information on the web, people visit web sites or
browse a large number of web pages related by key words
with the help of search engines. However, manually visiting
and searching sites is very time-consuming. Some
researchers, therefore, propose to integrate useful data over
the whole Internet with uniform schemes, so, people can
easily access and query the data with relational database
techniques. At the same time, the integrated data can be
mined to provide value-added services, such as comparison
shopping. As the data sources on the Internet are scattered
and heterogeneous, it is very difficult to integrate data from
web pages. On the other hand, web pages may present
information with embedded structured data, most of which
comes from backend relational database systems. Finding a
way to extract structured data from semi structured web
pages and integrating the data with uniform schemes is
necessary.
Web information extraction (WIE) is an important task for
information integration. Multiple web pages may present
same information using completely different formats or
syntaxes, which makes integration of information a
challenging task. Structure of current web pages is more
complicated than ever and is far different from their layouts
on web browsers. Due to the heterogeneity and lack of
structure, automated discovery of targeted information
becomes a complex task.A typical web page consistsofmany
blocks or areas, e.g., main content areas, navigation areas,
advertisements, etc. For a particular application, only part of
the information is useful, and the rest are noises. Hence, it is
useful to separate these areas automatically. Web
information extraction is concerned with the extraction of
relevant information from Web pages and transforming it
into a form suitable for computerized data-processing
applications. Example applications of WIE include: price
monitoring, market analysis and portal integration.
This paper is divided into several sections. Section 2
describes the related work on theoretical background and
Web information extraction In Section 3 we discuss our
proposed methodology in detail. Section 4 discusses the
result of our experimental tests while Section 5 concludes
this paper.
2. RELATED WORK
Information extraction from web pages is an active research
area. The existing works in Web data extraction can be
classiﬁed according to their automation degree (forasurvey,
see [5]). There are several approaches [4], [6], [9], [10], [15]
for structured data extraction, which is also called wrapper
generation. The first approach [9] is to manually write an
extraction program for each web site based on observed
format patterns of the site. This manual approach is very
labor intensive and time consuming. Hence, it does not scale
to a large number of sites. The second approach [10] is
wrapper induction or wrapper learning, which is currently
the main technique. Wrapper learning works as follows: The
user first manually labels a set of trained pages. A learning
system then generates rules from the training pages. The
resulting rules are then applied to extract target items from
web pages. These methods either require prior syntactic
knowledge or substantial manual efforts.
The third approach [4] is the automatic approach. The
structured data objects on a web are normally database
records retrieved from underlying web databases and
IJTSRD28010

International Journal of Trend in Scientific Research and Development (IJTSRD) @ www.ijtsrd.com eISSN: 2456-6470
displayed in webpages withsomefixedtemplates.Automatic
methodsaim to find patterns/grammars from the webpages
and then use them to extract data. Examples of automatic
systems are IEPAD[4],ROADRUNNER[6]extractsatemplate
by analyzinga pair of web pages of thesame class at a time. It
uses one page to derive an initial template and then tries to
match the second page with the template. Deriving of the
initial template has to be again done manually, which is a
major limitation of this approach.
Another problem with the existing automatic approaches is
their assumption that the relevant information of a data
record is contained in a contiguous segment of HTML code,
which is not always true. MDR [1] basically exploits the
regularities in the HTML tag structure directly. MDR works
well only for table and form enwrapped records while our
method does not have this limitation. MDR algorithm makes
use of the HTML tag tree of the web page to extract data
records from the page. However, an incorrecttagtreemaybe
constructed due to the misuse of HTML tags, which in turn
makes it impossibleto extract data records correctly. DEPTA
[14] uses visual information(locationsonthescreenatwhich
the tags are rendered) to find data records. Rather than
analyzing theHTML code, thevisual informationisutilizedto
infer the structural relationship among tags and to construct
a tag tree. But this method of constructing a tag tree has the
limitation that, the tag tree can be built correctly only as long
as the browser is able to render the page correctly.
Another similar system is Vints[8]. Vints proposes an
algorithm to find SRRs (search result records) fromreturned
pages of the search engines. However,ourmethodfocuseson
list pages of same presentation template. Although some
aspects and pieces of web information extraction may be
around in various techniques, the important of this paper
focus on the some interestingfeatures of web page andblock
clustering by appearance similarity.
3. SYSTEM OVERVIEW
This section describes the proposed automatic data records
extraction from web pages. Inthissystem,therearefivesteps
to extract data record from semi structured web page. The
algorithm for the proposed system proposed system is as
follows:
Algorithm: Data Records Extraction
1. Input :HTML Web page
2. Output :Extracted data table
3. Create DOM tree for input web page P and cleaning
useless nodes.
4. Segment the web page into several raw chunks/blocks
Bi={b1,b2…bi}.
5. Filter the noisy blocks based on heuristic rules.
6. Cluster the remaining blocks based on their appearance
similarity.
7. Labeling the data attributes for extracted data record.
Figure1 The Data Records Extraction Algorithm
First of all, input HTML page is changing DOM tree and
cleaning useless node as a preprocessing step. Secondly, we
tick out several raw chunks as a first round andthenfilterthe
noisy block based on noisy features of web pages. Then,
thirdly these blocks are clustered by proposed block
clustering method. Finally extract information from input
web page.
The goal of the system is to offer the extracted data records
from web page to the information integration systemsuchas
price comparison system and recommendation system.
3.1. Features in Web Pages
Web pages are used to publish information to users, similar
to other kinds of media, such as newspaper and TV. The
designers often associate different types of information with
distinct visual characteristics (such as font, position, etc.) to
make the information on Web pages easy to understand.Asa
result, visual features are important for identifying special
information on Web pages.
Position features (PFs).Thesefeaturesindicatethelocationof
the data region on a Web page.
PF1 : Data regions are always centered horizontally.
PF2 : The size of the data region is usually large relative to
the area size of the whole page.
Sincethe data records are thecontentsinfocusonwebpages,
web page designers always have the region containing the
data records centrally and conspicuously placed on pages to
capture the user’s attention. By investigating a large number
of web pages, first, data regions are always located in the
centre section horizontally on Web pages. Second, the size of
a data region is usually large when there are enough data
records in the data region. The actual size of a data region
may change greatly because it is not only influenced by the
number of data records retrieved, but also by what
information is included in each data record.
Figure2. Layout models of data records on web pages.
Layout features (LFs). These features indicate how the data
records in the data region are typically arranged.
LF1 : The data records are usually aligned flush left in the
data region.
LF2 : All data records are adjoining.
LF3 : Adjoining data records do not overlap, and the space
between any two adjoining records is the same.
Data records are usually presented in one of the two layout
models shown in Fig. 3. In Model 1, the data records are
arranged in a single column evenly, though they may be
different in width and height. LF1 implies that the data
records have the same distance to the left boundary of the
data region. In Model 2, data recordsarearrangedinmultiple
columns, and the data records in the same column have the
same distance to theleftboundaryofthedataregion.Because
most Web pages follow the first model, we only focus on the
first model in this paper. In addition, data records do not
overlap, which means that the regions of different data
records can be separated.

Appearancefeatures (AFs). These features capturethevisual
features within data records.
AF1 : Data records are very similar in their appearances,
and the similarity includes the number of the images
they contain and the fonts they use.
AF2 : The data items of the same semantic in different data
records have similar presentations with respect to
position, size (image data item), and font (text data
item).
AF3 : The neighboring text data items of different
semantics often use distinguishable fonts.
AF1 describes the visual similarity at the data record level.
Generally, there are three types of data contents in data
records, i.e., images, plain texts (the texts without hyper
links), and link texts (the texts with hyperlinks).AF2andAF3
describe the visual similarity at the data item level. The text
data items of the same semantic always use the same font,
and the image data items of the same semantic are often
similar in size. AF3 indicates that the neighboring text data
items of different semantics often use distinguishable fonts.
3.2. DOM Tree Generation and Clean useless node
(Preprocessing)
To begin with, a DOM tree should be generated from html
tags of the page. At the same time, features/attributes about
the node should be included in thesetreenodes,respectively;
described in next section.
Pre-processing is necessary in order to clean HTML pages,
e.g., to remove header details, scripts, styles, comments,
hidden tags,
space, tag properties, empty tags, etc. In this step, the white
relax function of thejsoup parsing tool for removingcleaning
HTML tag.First of all, we need to eliminate these nodestoget
clean Html page for further processing.
3.3. Filtering Noisy Block
Web page designers tend to organize their content in a
reasonable way: giving prominence to important things and
deemphasizing the unimportant parts with proper features
such as position, size, color, word, image, link, etc. All of
product page features are related to the importance. For
example, an advertisement may contain only images but no
texts, a contact information bar may contain email, and a
navigation bar may contain quite a few hyperlinks. However,
these features have to be normalized by the featurevalues of
the whole page to reflect the image of the whole page. For
example, the LinkNum of a blockshouldbenormalizedbythe
link numbers of the whole page. Then all these features are
formulated with equation (1).
𝑓𝑖(𝑎) =
number of attributes in block i
number of these attributes in whole page
(1)
Firstly, some conclusions are given on product features. All
those conclusionsareaccordingtotheobservationofproduct
list page on web site.
1. If a block contains email elements, then it is entirely
possible a contact block.
2. If TextLen/LinkTextLen<threshold, then it is quite
possible a hub block [7].
3. If <p> is included in a block, then this block is possible
authority block[7].
4. If the normalized LinkNum > threshold, then it is quite
possible a hub block.
Accordingly, these rules are calculated into equation (2) and
then F indicates the possibility of noisy block.
F=⅀∞i.fi(b)=α1.femail(b)+α2.ftextlen/linktexlen(b)+…+ α4.flinks(b), ⅀αi=1
(2)
Where∞i is coef icient, we can set different weightsonblock
importance respectively. Additionally, all these parameters
can be adjusted to adapt to different conditions. Finally
regarding product features, an important block is extracted
for further processing. Consequently, filtering noisy blocks
can decrease the complexity of web information extraction
through narrowing down the processing scope.
3.4. Blocks Clustering for Data Region Identification
The blocks in the data region are clustered based on their
appearance similarity. Since there are three kinds of
information in data records, i.e., images, plain text and link
text, the appearancesimilarityofblocksiscomputedfromthe
three aspects. For images, we care about the size; for plain
text and link text, we care about the shared fonts. Intuitively,
if two blocks aremoresimilar on imagesize,font,theyshould
be more similar in appearance. The appearance similarity
formula between two blocks b1 and b2 is given below:
sim(b1,b2)=Wi * simImg(b1,b2)+Wpt * simPT(b1,b2)+Wlt*
simLT(b1,b2) (3)
Where simImg(b1,b2), simPT(b1,b2), and simLT(b1,b2) are
the similarity based on image size , plain text , and link text.
Wi, Wpt, and Wlt arethe weights of these similarities.Table1
gives the formulas to compute the component similarities
and the weights in different cases.
Table1. The formulas of block appearance similarity and the weights in different cases
Formulas Descriptions
𝒔𝒊𝒎𝑰𝒎𝒈(𝒃𝟏, 𝒃𝟐) =
𝑴𝒊𝒏{𝒔𝒂𝒊(𝒃 𝟏), 𝒔𝒂𝒊(𝒃 𝟐)}
𝑴𝒂𝒙{𝒔𝒂𝒊(𝒃 𝟏), 𝒔𝒂𝒊(𝒃 𝟐)}
sai(b)is total number of images in block b.
sab(b) is the total number of block b.
fnpt(b) is the total number of fonts of the plain texts in block b.
sapt(b) is the total number of the plain texts in block b.
fnlt(b) is the total number of fonts of the link texts in block b.
salt(b) is the total number of the link text in block b.
𝑾𝒊 =
𝒔𝒂𝒊(𝒃 𝟏) + 𝒔𝒂𝒊(𝒃 𝟐)
𝒔𝒂 𝒃(𝒃 𝟏) + 𝒔𝒂 𝒃(𝒃 𝟐)
𝒔𝒊𝒎𝑷𝑻(𝒃𝟏, 𝒃𝟐) =
𝑴𝒊𝒏{𝒇𝒏 𝒑𝒕(𝒃 𝟏), 𝒇𝒏 𝒑𝒕(𝒃 𝟐)}
𝑴𝒂𝒙{𝒇𝒏 𝒑𝒕(𝒃 𝟏), 𝒇𝒏 𝒑𝒕(𝒃 𝟐)}
𝑾 𝒑𝒕 =
𝒔𝒂 𝒑𝒕(𝒃 𝟏) + 𝒔𝒂 𝒑𝒕(𝒃 𝟐)
𝒔𝒂 𝒃(𝒃 𝟏) + 𝒔𝒂 𝒃(𝒃 𝟐)
𝒔𝒊𝒎𝑳𝑻(𝒃𝟏, 𝒃𝟐) =
𝑴𝒊𝒏{𝒇𝒏𝒍𝒕(𝒃 𝟏), 𝒇𝒏𝒍𝒕(𝒃 𝟐)}
𝑴𝒂𝒙{𝒇𝒏𝒍𝒕(𝒃 𝟏), 𝒇𝒏𝒍𝒕(𝒃 𝟐)}
𝑾𝒍𝒕 =
𝒔𝒂𝒍𝒕(𝒃 𝟏) + 𝒔𝒂𝒍𝒕(𝒃 𝟐)
𝒔𝒂 𝒃(𝒃 𝟏) + 𝒔𝒂 𝒃(𝒃 𝟐)

Our block clustering method consists of two steps: The first
one is to build clusters by computing the similarity among
blocks. The similarity sim(b1,b2) between two blocks bi and
bj is computed by the equation (3).The second one is to
merge the resulting clusters. The threshold is trained from
sample page. So the cluster building procedure is simplified
as follows:
Procedure BlockClustering
Put all the blocks bi into the pool;
FOR(every block bi in pool){
compute the appearance similaritysim(bi,bj)bet:twoblocks
IF(sim(bi,bj) >threshold){
group bi and bj into a new cluster;
delete bi and bj from the pool;
}
ELSE{
create a new cluster for bi;
delete bi from the pool;
}
}
The second step is to merge clusters. To determine if two
clusters must be merged, we define the cluster similarity
simCkl between two clusters Ck and Cl as
the maximum value of sim(bi,bj), for every two blocks bi∈Ck
and bj∈Cl.
Procedure BlockMerging
FOR(every cluster Ck)
{
compute the simCkl with other clusters;
IF(simCkl >threshold){
clusters Ck and Cl are merged;
}
}
4. EXPERIMENTAL RESULTS
Our experiments were testing using commercial book store
web sites collected from different web site in Table 2. The
system takes as input raw HTML pages containing multiple
data records. The measure of our method are based on three
factors, the number of actualdata recordstobeextracted,the
number of extracted data records from the list page, and the
number of correct data records extracted from the list page.
Based on these three values, precision and recall are
calculated according to the formulas:
Recall=Correct/Actual*100
Precision=Correct/Extracted*100
According to abovemeasurement, wetested web pagesfrom
various book store web sites and check each page by
manually.
Table2. Results for selected Web Site
URL Precision Recall
gobookshopping.com 100 98
yahoo.com 98 97
allbooks4less.com 100 98
amazon.com 92.4 87.5
barnes&nobels.com 98 90
Average 97.68 94.1
Figure3. Result chart for selected web sites
5. CONCLUSION
In this paper, we have presented extraction information
content from semantic structure of HTML documents. It
relays on the observation that the appearance similarity of
data record in web page. Firstly, we segment a web page into
several raw chunks. Second, filter the noisy block. Then
proposed block clustering method groups remaining blocks
with their appearance similarity for data region
identification. Our method is automatic and it generates a
reliable and accurate wrapper for web data integration
purpose. In this case, neither prior knowledge of the input
HTML page nor any training set is required. We experiment
on multiple web sites to evaluate our method and the results
prove the approach to be promising.
Acknowledgment
I would like to greatly thank my supervisor, Dr. Thi Thi Soe
Nyunt, Professor of the University of Computer Studies,
Yangon, for her valuable advice, helpful comments, her
precious time and pointing me in the right direction. I also
want to thank Daw Aye Aye Khaing, Associate Professor and
Head of English Department andDawYuYuHlaing,Associate
Professor of English Department.
References
[1]. B Liu, R. Grossman and Y. Zhai, “Mining Data Records in
Web Pages”, ACM SIGKDD Conference, 2003.
[2]. B Liu and Y. Zhai, “NET – A System for Extracting Web
Data from Flat and Nested Data Records”, WISE
Conference, 2005.
[3]. Cai, D., Yu, S., Wen, J.-R. and Ma, W.-Y., VIPS: a vision-
based page segmentation algorithm, Microsoft
Technical Report.
[4]. Chang, C-H., Lui, S-L. “IEPAD: Information Extraction
Based on Pattern Discovery”, WWW-01, 2001.
[5]. Chang, C.-H., Kayed, M., Girgis, M., and Shaalan, K.
(2006). “A survey of web information extraction
systems”, IEEE Transactions on Knowledge and Data
Engineering, 18(10):1411–1428.
[6]. Crescenzi, V. and Mecca, G. “Automatic information
extraction from large websites”, Journal of the ACM,
2004, 51(5):731–779.
[7]. D. Cai, H. Xiaofei, W. Ji-Rong, and M. Wei-Ying, “Block-
level Link Analysis”, SIGIR'04, July 25-29, 2004.
80
82
84
86
88
90
92
94
96
98
100
Precision Recall

[8]. H. Zhao, W. Meng, Z. Wu, V. Raghavan, C. Yu, “Fully
Automatic Wrapper Generation for Search Engines”,
WWW Conference, 2005.
[9]. J. Hammer, H. Garcia Molina, J. Cho, and A. Crespo,
“Extracting semi-structuredinformationfromtheweb”,
In Proceeding of the Workshop on the Management of
Semi-structured Data, 1997.
[10]. Kushmerick, N, “Wrapper Induction: Efficiency and
Expressiveness. Artificial Intelligence”, 118:15-68,
2000.
[11]. M. Kayed, C.-H. Chang, “FiVaTech: Page-Level WebData
Extraction from Template Pages”, IEEE TKDE, vol. 22,
no. 2, pp. 249-263, Feb. 2010.
[12]. Shian-Hua Lin, Jan-Ming Ho, “Discovering Informative
Content Blocks from Web Documents”, IEEE
Transactions on KnowledgeandDataEngineering,page
41-45, Jan, 2004.
[13]. Yang, Y. and Zhang, H. “HTML page analysis based on
visual cues”, Z Niu, LiuLing Dai,YuMing Zhao,
“Extraction of Informative Blocks from web pages”, in
the Proceedings of International Conference on
Advanced Language Processing and Web Information
Technology, 2008.
[14]. YuJuan Cao, ZhenDong Niu, LiuLing Dai,YuMing Zhao,
“Extraction of Informative Blocks from web pages”, in
the Proceedings of International Conference on
Advanced Language Processing and Web Information
Technology, 2008.
[15]. Y. Zhai, and B. Liu, “Web Data Extraction Based on
Partial Tree Alignment”, WWW Conference, 2005.CA:
University Science, 1989.

The Data Records Extraction from Web Pages

More Related Content

What's hot (16)

Similar to The Data Records Extraction from Web Pages (20)

More from ijtsrd (20)

Recently uploaded (20)

The Data Records Extraction from Web Pages