Some of my XML/Internet Research Projects

1

Some of my XML/Internet

Research Projects

CSCI 6530October 5, 2005

Kwok-Bun YueUniversity of Houston-Clear Lake

10/5/2005 Bun Yue: [email protected], http://dcm.uhcl.edu/yue slide 2

Content

• Areas of My Research Interest• Some Current Projects• Storage of XML in Relational Database• Example Internet Computing Projects• Conclusions


Areas of My Research Interest

• Internet Computing• XML • Databases• Concurrent Programming


Content



Some Current Projects

• Storage of XML in relational database• Measuring Web bias using authorities and

hubs• Measuring information quality of Web pages• Distributed computer security laboratory• Collaborative Open Community for

developing educational resources• Generalized exchanges within organizations


Some Recent Student Work

• McDowell, A., Schmidt, C. & Yue, K., Analysis and Metrics of XML Schema, Proceedings of the 2004 International Conference on Software Engineering Research and Practice, pp 538-544, Las Vegas, June 2004.

• Yang A., Yue K., Liaw K., Collins G., Venkatraman J., Achar S., Sadasivam K., and Chen P., Distributed Computer Security Lab and Projects, Journal of Computing Sciences in Colleges. Volume 20, Issue 1. October 2004.

• Yue, K., Alakappan, S. and Cheung, W., A Framework of Inlining Algorithms for Mapping DTDs to Relational Schemas, Technical Report COMP-05-005, Computer Science Department, the Hong Kong Baptist University, 2005, http://www.comp.hkbu.edu.hk/en/research/?content=tech-reports.

http://www.comp.hkbu.edu.hk/en/research/?content=tech-reports




Content



Storing XML in RDB

• Advantages:– Mature database technologies.– May be queried by

• XML technology: e.g. XPath, XQuery.• RDB technology: e.g. SQL.

• Disadvantages: – impedance mismatch: XML and relations

are different data models.


Related Issues

• Effective mapping XML DTDs (~ ordered tree model) to relational schemas.

• Mapping of XML queries (e.g. XQuery) to RDB queries (e.g. SQL).

• Mapping of RDB query results back to XML format.


Related Work and Context

• Mapping – With or without schemas for XML.– With or without user input.

• Schemas for XML:– Document Type Definition (DTD)– XML Schema

• We consider mapping with DTD and without user input.


Naïve Mapping

• An XML element is mapped to a relation.

Example 1a:XML:

<a><b><c><d>hello</d></c></b></a>-> Relations: a, b, c and d.


Problems of Naïve Mapping

• Many relations.• Ineffective queries: multiple query joins.Example 1b:XPath Query: //aSQL Query: need to join the relations a, b,

c and d.


Inlining Algorithms

• First proposed by Shanmugasundaram, et. al.

• Expanded by Lu, Lee, Chu and others.• Extended in various directions by various

researchers, e.g.,– Preserving XML element orders.– Preserving XML constraints.

• Do not consider extensions here.


Basic Idea of Inlining Algorithms

• Inline child element into the relation for the parent element when appropriate.

• Different inlining algorithms differ in inlining criteria.

Example 1c: XML: <a><b><c><d>hello</d></c></b></a>

Inlined Relation: a.


Inlining Algorithms

• Child elements & attributes may be inlined.

• Child elements may not have their own relations.

• Results in less number of relations.• In general, more inlining -> less joins.


Inlining Algorithm Structure

1. Simplification of DTD.2. Generation of DTD graphs3. Generation of Relational Schemas


Our Preliminary Results

1. A more complete and optimal DTD Simplification Algorithm

2. A generic DTD Graph that can be used by inlining algorithms.

3. Inlining Considerations: framework for analyzing inlining algorithm

4. A new and aggressive inlining algorithm


Examples of Our Work

• Use DTD Simplification as an example of the flavor of our work.

• Show the new Inlining Algorithm.


Brief Introduction to DTD

• DTD: a simple language to describe XML vocabulary:– Element declarations: contents of

elements.– Attribute declarations: types and properties

of attributes.• DTD is still very popular.


DTD Element Declarations

• Define element contents:– #PCDATA: string– ANY: anything go– EMPTY: no content (attributes only)– Content models: child elements.– Mixed contents: child elements and strings.


DTD Example

Example 2: A complete DTD<!ELEMENT addressBook (person+)><!ELEMENT person (name,email*)><!ELEMENT name (last,first)><!ELEMENT first (#PCDATA)> <!ELEMENT last (#PCDATA)><!ELEMENT email (#PCDATA)><!ATTLIST person id ID #REQUIRED>


Operators for Element Declaration

• ,: sequence• +: 1 or more• *: 0 or more• ?: optional; 0 or 1• |: choice• (): parenthesis


Simplification of DTD

• Mapping of DTD to Relational Schemas:– Input: DTDs– Output: Relational Schemas

• DTD can be complicated => simplification.

Example 3:<!ELEMENT a (b,((b+,c)|(d,b*,c?)),

(e*,f)?)>


Simplification Principles

• The relational schema needs to store all possible scenarios.

• Some relations/columns may not be populated in some instances.

Example 3:<!ELEMENT a (b|c)> and<!ELEMENT a (b,c)>:May be the same from the RDB’s point of

view.


Simplification Details

• Comma-separated clauses: only operators remain: (), , and *.– + -> *, e.g. a+ -> a*.– Removal of | and ?, e.g. (a|b?) -> (a,b) – Removal of (), e.g. (a, (b)) -> (a,b)– Removal of repetition, e.g. (a, b, a) -> (a*, b)

• Note that element orders are not preserved.


Previous Simplification Results

• Not complete: e.g.– Shanmugasundaram: not specify how to

handle |.– Lu: not specify how to remove ().

• Not optimal (may generate * when it is not needed).

Example 4a: For Lu and Lee, 2 steps:(b|(b,c)) -> (b,b,c) -> (b*,c)


Our Simplification Algorithm

• A set of definitions.• A set of 7 simplification rules.• An algorithm on how and when to use

them.

Example 4b: For us, 1 step:(b|(b,c)) -> (b,c)


Simplification Rules


Simplification Algorithm


Complexity

Time complexity = O(Nop)

Where Nop is the total number of operators (including parentheses) in the element declarations of the DTD.


Advantages

• Complete: handle all DTDs.• Optimal: in the sense that * will not be

generated if not needed.

Example 5:<!ELEMENT a (b,((b+,c)|(d,b*,c?)),

(e*,f)?)> => <!ELEMENT a (b*,c,d,e*,f)>


A New Inlining Algorithm (1)

• Aggressive in inlining.• More complete. • Elaborated algorithms.• Handle more details: e.g. element types

of ANY, EMPTY and mixed contents.








Main Results

• Yue, K., Alakappan, S. and Cheung, W., A Framework of Inlining Algorithms for Mapping DTDs to Relational Schemas, Technical Report COMP-05-005, Computer Science Department, the Hong Kong Baptist University, 2005, http://www.comp.hkbu.edu.hk/en/research/?content=tech-reports.




Future Works

• Implemented the algorithms and tested with many DTDs.

• Need to implement the XQuery/SQL bridge for performance study.


Content



Measuring Web Bias

• Search engines dominate how information are accessed.

• Search results have major social, political and commercial consequences.

• Are search engines bias?• How bias are them?


Previous Works

• To measure bias, results should be compared to a norm.

• The norm may be from human experts.• Mowshowitz and Kawaguchi: the

average search result of a collection of popular search engines as the norm.


Mowshowitz and Kawaguchi

SE1

SEn

URLS1

URLSn

NORMURLS

URLVector1

URLVectorn

union NORMURL

Vector

Bias1

Biasn


Limitations

• Based on URL Vector -> cannot measure bias quality.


Our Approach

• Use Kleinberg’s HITS algorithm to create clusters, authorities and hubs of the result norm URLs.

• Use them as norm clusters, authorities and hubs.

• Measure distances between norms and individual results as bias.


HITS

• Obtain a directed graph G where– Node: page– Edge: URL link from between pages.

• Two indices: xp,i (authority) & yp,i (hub)• Iterate until steady state:

– xp,i+1 <- ∑ q,q->p yq,i

– yp,i+1 <- ∑ q,p->q xq,i


Our Approach

SE1

SEn

URLS1

URLSn

NORMURLS

URLVector1

URLVectorn

union NORMClusterVector

Bias1

Biasn

NORMCluster

ClusterVector1

ClusterVectorn


Current Progress

• Implemented previous results.• Implemented vector analysis• Implemented HITS algorithm, but it is not

accurate enough:– ‘Conglomerate’ effect.


Measuring Page’s Information Quality

• People find information from Web pages.• How good is the content of a given page?


Previous Works

• Measuring different kinds of quality:– Web site design quality– Navigational quality

• Many framework on how to measure information quality:– Most results in surveys so users can rank

informational quality.– Very few automated or semi-automated tool.


Our Objectives

• Build automated and/or semi-automated tool to measure and/or assist user to measure information quality of a Web page.


Approach

• Hypothesis, measure, usage guidelines.• Example:

– Hypothesis: a Web page with many spelling mistakes is likely to have low information quality.

– Measures: • Show frequencies of word occurrences.• Show percentage of spelling ‘mistakes’.

– Usage guideline:• Spelling ‘mistakes’ may not be actual mistakes (e.g.

UHCL).


Metrics

• Many potential metrics. Some examples:– Broken links– HTML Quality– Domain names– Page ranking and popularity– Appearance in directory structure– History (e.g. Way back machine)– Currency (e.g. last modified)– Author (e.g. Meta tag)


Current Progress

• ‘Pre-alpha’ prototype: http://dcm.cl.uh.edu/yue/util/pageInfo.pl

• A capstone project

http://dcm.cl.uh.edu/yue/util/pageInfo.pl


Content

• Areas of My Research Interest.• Some Current Projects• Storage of XML in Relational Database• Example Internet Computing Projects• Conclusions


Conclusions

• Good time to do applied computing research in the Web and XML areas.

• Style: hands-on supervision + publications.

• Don't forget to donate a scholarship to the School if your future research leads to a windfall.


Questions?

• Any Questions?• Thanks!

Documents

Some of my XML/Internet Research Projects