Building a Multi-level Database for Efficient Information Retrieval: A Framework Definition.
ABSTRACT With the explosive growth of the Internet and the World Wide Web, the amount of information available online is growing in an exponential manner. As the amount of information online constantly increases, it is becoming increasingly difficult and resource demanding to search and locate information in an efficient manner. Information overload has become a pressing research problem since current searching mechanisms, such as conventional search engines, suffer from both low- precision and low-recall. It is clear that a more dynamic, scalable and accurate searching methodology needs to be developed to overcome these limitations. This paper proposes a methodology consisting of an amalgamation of several research areas such as Web mining and relational database systems. We develop a proof of concept prototype which consists of an agent used to extract information from individual Web pages and a dynamic multi-level relational schema to encapsulate this information for later processing. The prototype provides users with a higher level of scalability and flexibility and can be utilized for searching the Internet and Intranets across large-scale organizations.
- SourceAvailable from: fiu.edu
Conference Paper: Web document classification based on fuzzy association[Show abstract] [Hide abstract]
ABSTRACT: In this paper, a method of automatically classifying web documents into a set of categories using the fuzzy association concept is proposed. Using the same word or vocabulary to describe different entities creates ambiguity, especially in the web environment where the user population is large. To solve this problem, fuzzy association is used to capture the relationships among different index terms or keywords in the documents, i.e., each pair of words has an associated value to distinguish itself from the others. Therefore, the ambiguity in word usage is avoided. Experiments using data sets collected from two web portals: Yahoo! and Open Directory Project are conducted. We compare our approach to the vector space model with the cosine coefficient. The results show that our approach yields higher accuracy compared to the vector space model.Computer Software and Applications Conference, 2002. COMPSAC 2002. Proceedings. 26th Annual International; 02/2002
Article: Databases deepen the Web[Show abstract] [Hide abstract]
ABSTRACT: The Web has become the preferred medium for many database applications, such as e-commerce and digital libraries. These applications store information in huge databases that users access, query, and update through the Web. Database-driven Web sites have their own interfaces and access forms for creating HTML pages on the fly. Web database technologies define the way that these forms can connect to and retrieve data from database servers. The number of database-driven Web sites is increasing exponentially, and each site is creating pages dynamically-pages that are hard for traditional search engines to reach. Such search engines crawl and index static HTML pages; they do not send queries to Web databases. The information hidden inside Web databases is called the "deep Web" in contrast to the "surface Web" that traditional search engines access easily. We expect deep Web search engines and technologies to improve rapidly and to dramatically affect how the Web is used by providing easy access to many more information resources.Computer 02/2004; · 1.68 Impact Factor
- [Show abstract] [Hide abstract]
ABSTRACT: In this paper, we present Google, a prototype of a large-scale search engine which makes heavy use of the structure present in hypertext. Google is designed to crawl and index the Web efficiently and produce much more satisfying search results than existing systems. The prototype with a full text and hyperlink database of at least 24 million pages is available at http://google.stanford.edu/ To engineer a search engine is a challenging task. Search engines index tens to hundreds of millions of web pages involving a comparable number of distinct terms. They answer tens of millions of queries every day. Despite the importance of large-scale search engines on the web, very little academic research has been done on them. Furthermore, due to rapid advance in technology and web proliferation, creating a web search engine today is very different from three years ago. This paper provides an in-depth description of our large-scale web search engine -- the first such detailed public description we know of to date. Apart from the problems of scaling traditional search techniques to data of this magnitude, there are new technical challenges involved with using the additional information present in hypertext to produce better search results. This paper addresses this question of how to build a practical largescale system which can exploit the additional information present in hypertext. Also we look at the problem of how to effectively deal with uncontrolled hypertext collections where anyone can publish anything they want. Keywords World Wide Web, Search Engines, Information Retrieval, PageRank, Google 1.Computer Networks. 01/1998; 30:107-117.
Building a multi-level database for efficient information retrieval:
A framework definition.
Spiridon C. Denaxas and Christos Tjortjis
Information Systems Group, School of Informatics
University of Manchester, PO Box 88, Manchester, M60 1QD, UK
S.Denaxas@postgrad.manchester.ac.uk | email@example.com
With the explosive growth of the Internet and the World
Wide Web, the amount of information available online is
growing in an exponential manner. As the amount of
information online constantly increases, it is becoming
increasingly difficult and resource demanding to search
and locate information in an efficient manner.
Information overload has become a pressing research
problem since current searching mechanisms, such as
conventional search engines, suffer from both low-
precision and low-recall. It is clear that a more dynamic,
scalable and accurate searching methodology needs to be
developed to overcome these limitations.
This paper proposes a methodology consisting of an
amalgamation of several research areas such as Web
mining and relational database systems. We develop a
proof of concept prototype which consists of an agent
used to extract information from individual Web pages
and a dynamic multi-level relational schema to
encapsulate this information for later processing. The
prototype provides users with a higher level of scalability
and flexibility and can be utilized for searching the
Internet and Intranets across large-scale organizations.
Information retrieval, World Wide Web, Semantic
Retrieval, Internet, Intranet.
The World Wide Web is potentially the world’s largest
knowledge base and can be viewed as one huge, diverse,
hybrid and dynamically distributed database. A recent
survey  revealed that the information available online
doubles every 18 months whereas the number of personal
homepages doubles every six months. Text and hypertext
are used for digital libraries, personal homepages,
reviews, news casts, product catalogues, newsgroups,
scientific and academic articles and medical reports for
individuals, organizations and projects.
The majority of documents that are available online can
be characterized as heterogeneous: they do not conform to
a consistent standard or style of authorship, nor such
standard exists . Web documents are authored by a
diverse set of individuals sharing different technical and
cultural backgrounds. Thus, the total volume of
heterogeneous text and hypertext data greatly exceeds that
of structured data.
Searching this vast volume of semi-structured
information that is available online poses a significant
challenge since it is more sophisticated and dynamic than
the information that any current database architecture can
store and manipulate. Taken as a whole, the set of Web
pages available lacks a unifying structure and presents a
far more complex authoring, style and context variation
than witnessed in traditional text archives . This
increased level of complexity renders “off-the-shelf”
database management, information
information searching solutions practically impossible to
use. It is clear that searching the Internet in an efficient
manner is a challenging field in which research needs to
be undertaken. This vast volume of accessible information
online has raised many new opportunities and challenges
for knowledge discovery
In this paper, we provide an overview of the current
problems associated with the online information searching
domain, the current trends and existing solutions and offer
a detailed framework specification of a prototype
designed to overcome of these limitation and provide
users with the ability to perform more accurate and
The paper’s structure is as follows: In section 2 we
discuss the dominant methods for locating information
online and illustrate their current limitations. Section 3
depicts the proposed methodology and a detailed analysis
of its core components is provided. A prototype system is
discussed and assessed in section 4 and finally
conclusions and directions for future work are provided in
and data engineering
2. Background Issues
Users obtain information on the Internet or individual
Intranets using two dominant procedures :
a) manually browsing b) utilizing a Web search engines
These procedures are briefly discussed and evaluated
2.1 Manual browsing
Browsing refers to the act of following hyperlinks
between Web sites and traversing through them until the
information requested is located. This process is very
often tentative and unsatisfactory since a user might be
forced to spend large amounts of time in order to
successfully locate the
Additionally, a recent survey  revealed that the deep
Web holds approximately 500 billion Web pages in it and
that public information hidden in it is 400 to 500 times
larger than what users can access through the surface
Web. This hidden information is only accessible by
performing intelligent queries to various distributed
database systems. It becomes clear that given the
Internet’s ever-growing nature, manual browsing will
soon become an extremely resource inefficient procedure
for locating information online.
2.2 Search engines
The second dominant method for locating information
online is by using Web search engines. According to the
manner in which information is collected and indexed,
Web search engines can be divided into three main
categories: directory search engines, crawler-based search
engines and meta-search engines .
Directory search engines are engines which consist of
thematic directories a user can browse and locate
information. These directories are manually built and only
represent a very small percentage of the Word Wide Web
since their respective database cannot be updated in real
time to synchronize with the Internets volatile nature.
Crawler-based search engines are engines which make
use of correlated programs called Web spiders in order to
traverse an individual Web site and its respective pages
analyze their content and finally add them to a large index
database automatically. Finally, meta-search engines are
engines that make use of various ranking algorithms such
as PageRank ,, or HITS  in order to calculate
some degree of authority in the results obtained, and
effectively rank them before returning them to the users.
Conventional search engines have a number of
problems and inherent limitations associated with them.
Given the World Wide Web’s volatile and expanding
nature, most search engines have a very limited coverage
of the Internet. One could argue that it is becoming
practically impossible to maintain an up-to-date index of
the Internet for searching. Additionally, conventional
search engines offer little to none interaction with the
user: a user cannot make detailed specifications apart
from entering a number of keywords.
Search engines are also known  to suffer from poor
accuracy: they have both low recall (fraction of relevant
documents that are retrieved) and low precision (fraction
of retrieved documents that are relevant).
A typical search would return a very large amount of low-
accuracy results which the user has to manually browse in
order to locate the information requested.
Common text searching issues such as synonymy,
polysemy (occurring when a word has more than one
meaning) and context sensitivity also become severe on
the Web ,. Finally search engines are also limited
by various secondary factors such as processing delays
and bandwidth bottlenecks.
It becomes clear that existing solutions for locating
information on the Internet and individual Intranets have
significant deficiencies with respect to robustness,
flexibility and precision .
3. Proposed Methodology
The proposed methodology consists of two individual
components developed separately: the agent and the
underlying relational database. The former is responsible
for identifying and extracting information from individual
Web pages it traverses whereas the latter is concerned
with storing and manipulating these data.
The agent is responsible for recursively traversing
individual Web sites and the respective Web pages that
are linked to them. While traversing a Web page, the
agent identifies and extracts a set of predefined elements,
processes them and forwards the final set of data to the
underlying relational database.
The agent extracts both semantic and descriptive
information from the individual Web pages it recursively
traverses. Semantic information can be defined as the
elements that capture an individual Web pages contents,
such as the number of frame elements, ordered or
unordered lists and image files on it. The agent is
responsible for correctly identifying the HTML meta-tags
composing each Web page, filtering out any invalid or
erroneous elements and enumerating each of the
predefined document elements. The agent processes data
by performing a number of sequential conceptual
“passes”. The elements the agent component is
responsible for extracting along with their respective
HTML meta-tags are summarized in table 3.1.
The agent heavily relies on proper syntax of the Hyper
Text Markup Language it encounters. Due to the Web’s
hybrid nature, syntax errors and inconsistencies due to
different coding styles are bound to exist. While these will
not generate fatal errors, a probability that inconsistent or
incorrect data will be inserted into the relational database
<IMG SRC= “path”>
<A HREF = “path”>
<INPUT type= “checkbox”
Table 3.1: The predefined document elements the
Additionally, the agent will further process the number
of image media files it identifies on an individual Web
page and enumerates their respective types and sizes on a
separate process. The types currently supported from the
agent component are summarized in table 3.2.
Image File Description
Monochrome, 16 color, 256 color and
24-bit Bitmap image files
JPEG image file
Jpg, jpeg, jpe,
GIF image file
TIFF image file
PNG image file
Table 3.2: The supported image formats.
We define descriptive information as the set of
information the agent will extract which will provide the
user with an abstraction of the individual Web page
concerned. Descriptive information includes but is not
limited to the documents author, the character set and
codepage used or the Web pages title and language. The
descriptive information the agent component is
responsible for extracting is summarised in table 3.3.
category The document’s thematic category.
rank Specifies the document’s rank.
author The document’s author.
title The document’s title.
timestamp Denotes the time the document was
inserted into the system.
charset The document’s character set.
content The document’s content type.
language The document’s language code.
Table 3.3: Descriptive information extracted by the
To avoid data duplication and promote data consistency
and accuracy, filtering functions are included within the
core of the agent component. The agent downloads the
source code of a particular Web page and scans it for the
set of predefined elements. Additionally, it also locates
and extracts hyperlinks pointing to other pages and places
them in a queue; the individual pages are considered
subsequently in queue order resulting in a breath fist
search. Links pointing to
advertisements or other non-relative information are
rejected. Finally, a visited URL hash exists which prevents
the agent from re-visiting the same Web page.
3.3 Relational database
The relational database component is responsible for
storing and manipulating the information the agent
identifies and extracts from the individual Web pages it
traverses. A dynamic schema was developed in order to
cope with the Web’s hybrid and dynamic nature.
Essentially, the relational database does not hold complete
Web pages, something extremely resource demanding and
unrealistic, but instead stores document abstractions
which encapsulate all the necessary information for latter
processing and querying.
The database is organized in a number of logical levels,
each of which holds a different type of information
concerning an individual Web page the agent component
Level 0: The World Wide Web it self, un-
altered. Although outside the context of the
system, it provides the essential base on which
the next logical levels are built upon.
Level 1: The semantic information the agent
extracts from the individual Web pages it
traverses. Layer 1 also contains a more detailed
analysis of the image media types the agent
Level 2: This level contains the descriptive
information the agent extracts from the
individual Web pages it traverses.
Following the Object Exchange Model (OEM) 
paradigm, each document abstraction is assigned with a
unique Object Identifier (OID) value automatically from
the relational database management system. This will
effectively form the binding between the information
concerning a particular Web page as it is distributed
amongst the different levels of the relational database.
external Web sites,
Finally, a number of predefined default values are used
when the agent fails to locate or extract an element from a
Web page; this promotes data integrity and consistency
and minimizes the need for data pre-processing when
applying data mining algorithms to the data harvested.
4. A prototype system
In order to demonstrate the proposed framework’s
feasibility and to perform some initial experiments, a
prototype system was developed equipped with the
majority of the features mentioned in this paper. As this
paper presents ongoing work, the systems specifications
constantly evolve and new specifications are defined.
The Web site chosen to be processed is the University
of Manchester: Institute of Science and Technology
(UMIST)  homepage and an internal limit of 500 Web
pages was set in order to truncate the output. The total
time taken to process the sample was: 02’:01” and a total
of 434 records were created. Out of the 434 records
processed, 430 were unique which translates to a mere
0.91% of duplicate information. The remaining records
where lost due to throttling occurring from the HTTP
The results obtained are summarized in tables 4.1 and
Table 4.1: The total number of elements processed by
Data Transfer Download
Table 4.2: The results in terms of data transfer.
0.45 MB 10.90 MB
236.0 kB/sec 9.9 kB/sec
92.3 kB/sec 3.8 kB/sec
The total size of the relational database datafiles after the
experiment is 1.28MB (1,352,126 bytes) which is
relatively small in contrast with the 11MB of data the
agent downloaded while processing the sample.
The data harvested by the agent can now be queried
using standard relational database query languages like
SQL. Additionally, one of the strengths of the defined
framework is its compatibility with existing data mining
tools and algorithms with no need of extensive
preprocessing; the default values used by the system
ensure data within the relational database remains
consistent. Using clustering, which aims in grouping
records together based on their similarity, hidden but
potentially useful patterns can be discovered and utilized
in the demanding field of Web content mining .
4.1 Case studies
In order to fully depict the defined framework’s
usability and flexibility, a number of case studies were
constructed illustrating the different queries that can be
performed on the data harvested by the agent.
4.1.1 First case study
A user is compiling a report on various statistical data
found on a Web site; the majority of the desired numerical
data on that particular Web site is located within table
elements. He/she does not require image files of any type
or any other data in the form of text. By making use of the
fields provided within the relational database, the user is
able to specify the minimum number of table elements a
Web page should have when the results are returned. The
system provides the user with the flexibility to exactly
specify the document elements he/she wishes to receive.
Furthermore, should the user require information that is of
a certain age, he/she can additionally specify the
minimum timestamp a document must have before it is
4.1.2 Second case study
A user is compiling a medical report on a particular skin
disease and wishes to obtain a number of images. He does
not require any type of numerical or textual data of any
form. Additionally, having to cope with a number of
standards, he only wishes to locate JPG type images
whose size does not exceed 150KB. By utilizing the fields
defined within the relational database component, the user
is able to exactly specify the minimum number of images,
their respective formats and sizes he wishes to receive.
The prototype system will use a standard query language
to search the data harvested by the agent component and
return it to the user.
It becomes clear by the case studies discussed above
that the defined framework provides the user with a much
greater level of flexibility than conventional and existing
search methods. Additionally, the proposed framework
can be used to efficiently locate information on a
corporate Intranet, giving users the ability to specify more
advanced query parameters than conventional search
5. Conclusion and future work
The amount of available information online is
exponentially growing, making efficient information
searching and retrieval mechanisms a pressing and
research-demanding issue. Existing techniques have
significant deficiencies with respect to robustness,
flexibility and precision. The current information location
trends, such as manual browsing and conventional search
engines were analyzed and their shortcomings discussed.
This paper investigated the area of locating, identifying
and extracting information available online with the
ultimate purpose of defining a system
which will eventually enable more sophisticated,
advanced and accurate queries to take place.
One could claim that this work only covers a rather
narrow scope of the information searching and knowledge
discovery domains. However, the main contribution of
this paper is the definition of a framework which can act
as the foundation on which more complex and advanced
systems can be built upon. The components themselves
can be integrated into larger systems and effectively assist
users into locating the requested information in an
accurate and efficient manner. Finally, data mining
algorithms, such as clustering and association analysis,
can be performed on the data harvested by the agent
component in order to reveal potentially useful but hidden
information on the basis of Web mining.
5.1 Future Work
We consider the following various alternatives in order
to enhance the proposed framework.
a) Automatic classification of documents.
In order to provide the user with a greater level of
flexibility, a thematic classification algorithm could be
integrated to the agent and utilized while traversing Web
pages. Documents can be classified into several key-topic
categories such as sports, economy-related, academic,
leisure, personal etc.. Users would be able to specify the
category they wish to receive results from thus effectively
narrowing down the amount of irrelevant topics returned
Additionally, a ranking algorithm could be applied on
the results before they are returned to the user. By
utilizing a ranking algorithm, such as PageRank, a certain
degree of authority would be calculated and displayed to
the user. This would enable him to locate the desired
information in a more efficient manner while increasing
the relevant quality of the information he receives from
 C. Yang, J. Hen, & H. Chen, Intelligent internet
searching agent based on hybrid simulated annealing,
Elsevier Decision Support Systems 28, 2000, 269-277.
 S. Chakrabarti, Data mining for hypertext: A tutorial
survey, SIGKDD Explorations 1(2), 2000, 1-11.
 Y. Li, X. Chen & B. Yang, Research on Web mining-
based intelligent search engine, Proc 1st IEEE Conf. on
Machine Learning and Cybernetics, Beijing, 2002, 386-
 T. Ghanem, W. Aref, Databases deepen the Web,
IEEE Computer, 2004, 116-117.
 C. Hsinchun, C. Yi-Ming, M. Ramsey, C. Yang, An
intelligent personal spider (agent) for dynamic Internet /
Intranet searching, Elsevier Decision Support Systems 23,
 S. Brin, L. Page, The anatomy of large-scale
hypertextual Web search engine, Elsevier Computer
Networks and ISDN Systems 30(1-7), 1997, FP11.
 L. Longzhauang, Y. Shang, W. Zhang, Link analysis,
improvement of HITS-based algorithms on Web
documents, Proc. 11th Elsevier International Word Wide
Web Conf., May 2002.
 R. Kosala, H. Blockeel, Web mining research, A
survey, SIGKDD Explorations 2(1), July 2000, 1-15.
 J. Pokorny, Web searching and information retrieval,
IEEE Computing in science and engineering 6(4),
July/August 2004, 43-48.
 K. Wang, H. Liu, Schema discovery for
semistructured data, 3rd International conf. on Knowledge
Discovery and Data Mining (KDD-97), 1997.
 The University of Manchester: Institute of Science
and Technology homepage, http://www.umist.ac.uk,
accessed on 05/08/2004.
 L. Liu, J. Chen, H. Song, The research of Web
Mining, Proc. 4th World Congress on Intelligent Control
and Automation, Shanghai, China, 2002, 2333-2337.
 C. Haruechaiyasak, M. Shyu, S. Chen, Web
document classification based on fuzzy association, Proc.
26th IEEE Annual International Computer Software and
Applications Conference (COMPSAC’02),2002,
 R. Tamir, R. Rapp, Mining the Web to discover the
meanings of an ambiguous word, Proc. 3rd International
IEEE Conf. on Data Mining (ICDM’03),2003.