1 INTRODUCTION

XML and related technologies are used extensively

for storing, managing, and exchanging information.

Our work is motivated by the consideration of docu-

ments describing requirements or business rules to be

met to achieve some designation or status. According

to (Business Rules Group 2000, Ross 1997), a busi-

ness rule is a statement that defines or constrains

some aspect of the business and is used to control the

behavior of the business. For example, in a university

calendar, we have rules that specify the valid collec-

tion of courses for a student to obtain a certain uni-

versity degree.

As shown in (McFadyen et al 2005), such rules

can be organized into a special structure, the so-

called synthesized query tree (SQT), to govern their

execution. An observation shows that such trees may

heavily overlap. Then, the separate storage of SQTs

not only leads to redundant space, but also causes

repeated evaluation of rules. For this reason, we frag-

ment individual SQTs and organize the sub-SQTs

into a hyper tree structure, in which a leaf node can

be a simple query or a subtree itself. This kind of

organization is similar to document fragmentation

(Salminen and Tompa 2001, Quint and Vatton 2004).

However, in our case, a fragment is a subtree repre-

senting a set of rules, which produces a piece of a

document dynamically. Our method also shares the

flavour of active XML documents (Abiteboul et al

2002, Abiteboul et al 2003, Bonifati et al 2001), by

which parts of the contents are generated by invoking

a program or a web service; but differs from these in

that we are concerned with the processing of business

rules, which are evaluated along a tree structure in a

bottom-up way.

The remainder of this paper is organized as fol-

lows. Next in section 2, we present the background

information on requirement documents and synthe-

sized query trees. In section 3, we present the SQT

fragmentation based on the so-called virtual SQTs,

which enable us to efficiently evaluate queries. Sec-

tion 4 presents a short conclusion and directions for

further work.

RULE-BASED QUERY TREE EVALUATION OVER

FRAGMENTED XML DOCUMENTS

Ron G. McFadyen

, Yangjun Chen

Department of Applied Computer Science, University of Winnipeg,

515 Portage Avenue, Winnipeg, Manitoba, Canada, R3B 2E9

Supported by NSERC 105709-03 (139988) (Natural Sciences and Engineering Council of Canada)

Supported by NSERC 239074-01 (242523) (Natural Sciences and Engineering Council of Canada)

Keywords: XML, business rule, query tree, query evaluation, document fragmentation and assembly.

Abstract: XML documents are used to hold information and to make exchanges between systems. In this paper, we con-

sider documents that embed knowledge and rules, which may contain considerable redundancy. To control

redundancy and to provide for efficient execution of queries, documents are decomposed into fragments that

are stored separately. Then, to materialize documents for end-users, they need to be dynamically constructed

from their sources (separately stored fragments) by evaluating rules, which requires database queries to be

executed according to the document structure.

199

G. McFadyen R. and Chen Y. (2006).

RULE-BASED QUERY TREE EVALUATION OVER FRAGMENTED XML DOCUMENTS.

In Proceedings of WEBIST 2006 - Second International Conference on Web Information Systems and Technologies - Internet Technology / Web

Interface and Applications, pages 199-205

DOI: 10.5220/0001237501990205

 SciTePress

2 REQUIREMENTS DOCUMENT

AND SQT STRUCTURE

In Figure 1, we show some typical majors found in a

university calendar. Their XML representation is

illustrated in Figure 2. In Figure 1, each document is

a requirement for graduation from a certain major.

For instance, in Figure 1(a), we specify that to

graduate with a 3-Year BSc in Geography a student

must satisfy all of: Graduation Requirement (com-

pletion of 90 credit hours), Residence Requirement

(completion of a minimum of 30 credit hours at the

university and a minimum of 18 credit hours in

Geography at the university), a General Requirement

(12 credit hours in Humanities subjects and 6 credit

hours in Science subjects), and a Major Requirement

for certain courses in Geography. The fact that all

four of these requirements must be met simultane-

ously is indicated by the attribute: combin-

ing=”AND”, associated with the element

GeographyRule in the second line of the document

shown in Figure 2.

As discussed in (McFadyen et al 2005), we can

construct a synthesized query tree over such a docu-

ment. (McFadyen et al 2005) presents two such tree

structures: the boolean and the general SQT; for sim-

plicity, here we only discuss the boolean SQT.

Definition 1: a boolean synthesized query tree

(BSQT) is a tree where each leaf node v is associated

3-Year BSc (Geography)

Graduation Requirement

90 credit hours

Residence Requirement

Degree: minimum 30 credit hours

Major: minimum 18 credit hours

General Degree Requirement

Humanities: 12 credit hours

Science: 6 credit hours

Major Requirement

Minimum 30 credit hours

Maximum 48 credit hours

Required Courses

23.202 Intro Geography I

23.203 Intro Geography II

23.331 Advanced Geography

Choice

23.205 Atmos Sci or 23.206 Earth Sci

a) Graduation Requirements - Geography

3-Year BSc (Physics)

Graduation Requirement

90 credit hours

Residence Requirement

Degree: minimum 30 credit hours

Major: minimum 18 credit hours

General Degree Requirement

Humanities: 12 credit hours

Science: 6 credit hours

Major Requirement

Minimum 36 credit hours

Maximum 54 credit hours

Required Courses

44.101 Intro Physics

44.203 Mechanics

44.331 Relativity

b) Graduation Requirements - Physics

Figure 1: Graduation Requirements.

Figure 2: XML for 3-Year BSc (Geography).

<GeographyRule title=“Degree Requirement for 3-Year BSc(Geography

)”

combining=”AND”>

<GraduationRule title=”Graduation Requirement”

display=”90 credit hours” query= “... ” ...>

</GraduationRule>

<ResidenceRule title = ”Residence Requirement”

combining = ”AND” >

<DegreeRule title = ”Degree”

display = ”minimum 30 credit hours”

query = ”SELECT sum(creditHours)

FROM studentHistory

WHERE studentNum = x

HAVING sum(creditHours) > 30” >

</DegreeRule>

<MajorRule title=”Major” display=”minimum 18 credit hours”

query=”... ” ...>

</MajorRule>

</ResidenceRule>

<GeneralRule title=”General Degree Requirement”

combining=”AND”>

<HumanitiesRule title=”Humanities”

display=”12 credit hours” query=”... ” ...>

</HumanitiesRule>

<ScienceRule title=”Science” display=”6 credit hours” query=”... “..

</ScienceRule>

</GeneralRule>

<MajorRule> title=”Major Requirement”

combining=”AND”>

<MinMaxRule

display=”Minimum 30 credit hours, Maximum 48 credit hours”

query=”... “ ...>

</MinMaxRule>

<ReqCoursesRule title=”Required Courses”

combining=”AND”>

<Course

display=”23.202 Intro Geography I” query=”...” ...> </Course>

<Course

display=”23.203 Intro Geography II” query=”...” ...> </Course

<Course

display=”23.331 Advanced Geography” query=”...” ...> </Cours

</ReqCoursesRule>

<ChoiceRule title=”Choice”

display=”23.205 Atmos Sci or 23.206 Earth Sci”

combining=”OR”>

</ChoiceRule>

</MajorRule>

</GeographyRule>

GROUP BY studentNum

and institution=”UW”

WEBIST 2006 - INTERNET TECHNOLOGY

200

with a boolean query , and each internal node v

is labelled with a tag T(v), and an operator θ = ∨ or ∧;

and each node v is assigned a boolean value, V(v),

determined as follows:

a) for a leaf node, V(v) is true if the return value

of is not empty; otherwise, it is false, and

b) for an internal node, with children v

, ... v

For instance, for the graduation requirement of

Geography, we will construct a tree structure as

shown in Figure 3, which represents the rules, que-

ries and relationships corresponding to the require-

ments shown in Figure 2. One of the defining

characteristics of a query tree is that database queries

are only present in leaf nodes.

To determine if a student can graduate, it is nec-

essary to evaluate the appropriate SQT and its que-

ries in the context of the student. Thus, to do this for

a number of students, we have to traverse the SQT

trees repeatedly, each time for a single student.

An observation of the queries present in the

leaves shows that a slight modification will facilitate

set processing. To see this, let’s have a look at the

query Q

shown in Figure 3. If we remove the condi-

tion studentNum = x from the where-clause, the exe-

cution of the query will find all the students with

grade point ≥ 1 and more than 90 credit hours.

Now, for a given set of students, to check whether

they are eligible to get a degree, we traverse the

query tree bottom-up. During this process, all the

queries attached with the leaf nodes are evaluated

against the student records and the results are trans-

ferred to the internal nodes for further checking the

specified logic conditions.

For students of a different major, a different SQT

will be instantiated and traversed. Obviously, if two

SQTs share a common subtree, this subtree will be

traversed two times and its queries executed twice.

For example, the General requirement in the docu-

ment shown in Figure 1(a) is completely the same as

the General requirement shown in Figure 1(b). If we

evaluate the General subtree on its own it is traversed

once and its queries are executed once. The returned

result is then separated according to student majors

and transferred to the parent in the respective SQTs.

In this way, the common subtree is evaluated only

once. This observation leads to the SQT fragments

discussed in the next section.

()

()θ…θ

()

Residence

Degree

Major

General

Major

Required

Courses

Huma-

Science

Graduation

or-node:

Choice

Complete 3-Year BSc (Geography)

and-node:

Figure 3: A Boolean SQT for graduation requirements.

select *

from StudentHistory

where studentNum =

and

gradePoint >= 1

group by studentNum

having sum(crHours >=90)

select *

from StudentHistory

where studentNum =

and

institution = ‘UW’

group by studentNum

having sum(crHours >=30)

select *

from StudentHistory

where studentNum =

and

major = ‘Geography’

group by studentNum

having sum(crHours >=18)

select *

from StudentHistory

where studentNum =

and

area = ‘humanities’

group by studentNum

having sum(crHours >=12)

select *

from StudentHistory

where studentNum =

and

area = ‘science’

group by studentNum

having sum(crHou rs >=6)

select *

from StudentHistory

where studentNum =

and

courseNum = 23.205

select *

from StudentHistory

where studentNum =

and

courseNum = 23.206

select *

from StudentHistory

where studentNum =

and

courseNum = 23.202

select *

from StudentHistory

where studentNum =

and

courseNum = 23.203

select *

from StudentHistory

where studentNum =

and

courseNum = 23.331

select *

from StudentHistory

where studentNum =

and

major = ‘Geography’

group by studentNum

having

sum(30 <= crHours <= 48)

nities

group by studentNum

RULE-BASED QUERY TREE EVALUATION OVER FRAGMENTED XML DOCUMENTS

201

3 DOCUMENTS:

FRAGMENTATION AND

EVALUATION

In this section, we discuss SQT fragmentation. First,

we show how to fragment SQTs in 3.1. Then, we dis-

cuss how to evaluate fragmented SQTs in 3.2.

3.1 SQT Fragmentation

In (McFadyen et al 2005), two kinds of SQTs:

boolean SQTs and general SQTs are defined. Both

can be fragmented to speed up query evaluation. For

simplicity, however, we show only how to fragment

the boolean SQTs and the general SQTs can be han-

dled in a similar way.

First, we introduce the concept of virtual boolean

synthesized query trees, based on which the boolean

SQT fragmentation is conducted.

Definition 2: a virtual boolean synthesized query

tree (VBSQT) is a tree where a leaf node v is either

a) associated with a boolean query , or

b) specifies a fragment that is another VBSQT

(such a leaf node is called a virtual leaf node),

and each internal node v is labelled with a tag T(v),

and an operator θ = ∨ or ∧; and each node v is

assigned a boolean value, V(v), determined as fol-

lows:

a) for a leaf node that is a query, V(v) is true if the

return value of is not empty; otherwise, it

is false, and

b) for a leaf node that specifies a fragment, V(v) is

the value of the fragment, and

c) for an internal node, with children v

, ... v

For instance, the tree shown in Figure 4(a) is a

virtual version of the tree shown in Figure 3, in

which the leaf nodes labelled with v

, v

, and v

rep-

resent the three trees shown in Figure 4(b), respec-

tively. They are singled out since they also belong to

other SQTs. To see this, examine the tree shown in

Figure 4(c), which is a VBSQT for the 3-Year BSc in

Physics and where v

, v

, and v

are three of its leaf

nodes, too.

When more than two SQTs are involved, more

complicated SQT fragmentation has to be considered

as illustrated in Figure 5.

In this figure, we show three SQTs: T

, T

, and

. Among them, T

and T

share a common subtree

T’; and T

and T

share a different common subtree

T’’. Furthermore, T’’ itself is a subtree of T’. In such

a case, we will generate five VBSQTs. They are T

T’ (which represents the tree obtained by replacing

T’ with a virtual leaf node in T

), T

/T’, T

/T’’, T’/

T’’, and T’’.

()

()θ…θ

()

Major

Required

Courses

Choice

Complete 3-Year BSc (Geography)

Graduatio

Residence

Degre

Major

General

Huma-

Science

nities

(a)

(b)

Major

Required

Courses

Complete 3-Year BSc (Physics)

(c)

Figure 4: Illustration for SQT fragmentation.

Residence

Major

T’

T’’

’’

Figure 5: Fragmentation of three SQTs.

WEBIST 2006 - INTERNET TECHNOLOGY

202

As an example, consider a 3-Year BA (English)

major where the Humanities requirement is speci-

fied, but no specification for the Science require-

ment. Thus, the subtree representing the Humanities

requirement in the SQT for the English major is a

proper subtree of the General requirement in the

Geography major as illustrated in Figure 6(a), in

which v

represents the subtree shown in Figure 6(b).

Accordingly, the virtual SQT for the English major

will be of the form shown in Figure 7.

Note that the situation gets more complicated if

we allow for some other major that specifies the Sci-

ence requirement but not the Humanities. In such a

case, the General requirement would then have two

subtrees common to some different SQTs.

In general, we have the following algorithm to

fragment any number of SQTs. The algorithm is fol-

lowed by an example.

Algorithm SQT-fragmentation

1. Let T

, T

, ..., T

be SQT trees;

2. Let , ..., be all the subtrees shared by T

and

≠

j);

3. Repeat until fragments have been created for all

≠

j).

a) From unmarked subtrees, select and mark if

for all , (i,j

≠

s,t). Generate a frag-

ment for . Mark any , if = .

b) For each , if , and there is no any

other such that , replace by

4. Generate VBSQTs: T

/{all }, ..., T

/{all }.

To explain the SQT-fragmentation algorithm we

consider a more complicated scenario involving

Geography Physics, and English, which are consid-

ered as T

, T

respectively in Step 1. Step 2 deter-

mines 9 common subtrees as shown below:

For Geography and Physics:

= Graduation subtree

= Degree subtree

= General subtree

For Geography and English:

= Graduation subtree

= Degree subtree

= Humanities subtree

For Physics and English:

= Graduation subtree

= Degree subtree

= Humanities subtree

In Step 3, we first generate 3 fragments for Grad-

uation requirement (= = ), Degree

requirement (= = ), and Humanities

requirement (= ) as these do not contain any

identified subtrees (see (a) in Step 3). Then, the frag-

ment for / will be created (see (b) in Step 3).

Finally, Step 4 generates the following VBSQTs: T

{ ∪ ∪ }, T

/{ ∪ ∪ }, T

{ ∪ ∪ }.

The fragmentation algorithm creates fragments

that, for a given set of documents, controls redun-

dancy present in rules by extracting common rules

into separate documents. Dividing a collection of

documents into sub-documents where those sub-doc-

uments are common components is a way of structur-

General

Science

Figure 6: Humanities as a fragment of General.

Humanities

(a) (b)

⊄Δ

⊃

Δ⊃

⊃Δ

Complete 3-Year BA (English)

Figure 7: VBSQT for English.

Major

Required

Courses

Choice

Residence

Major

RULE-BASED QUERY TREE EVALUATION OVER FRAGMENTED XML DOCUMENTS

203

ing documents into manageable pieces that can be

considered separately or in combination.

Figure 8 presents the 3-Year BSc Geography doc-

ument as presented in Figures 4(a) and 4(b). Note the

use of the Xinclude feature of XML (XML.org 2005)

to link a pair of documents.

The next section discusses the evaluation of these

documents which requires the documents be re-

assembled in some way.

3.2 Evaluation of Fragmented SQTs

To determine the students who can graduate, we need

to evaluate the necessary SQTs for the students in

question. If we were to follow the procedure in

(McFadyen et al 2005) we would instantiate all SQTs

for each degree and every common fragment would

be evaluated many times. In contrast, we present

another more efficient procedure based on a frag-

ment graph for evaluating graduation status of stu-

dents.

In order to evaluate a fragmented SQT, we con-

struct a directed graph, called a fragment graph, in

which each node represents a fragment (or say, a

VBSQT), and we have an edge from a node frag

another node frag

if frag

⊂ frag

and there is not

any node frag

such that frag

⊂ frag

. For

instance, the VBSQTs shown in Figures 4, 6 and 7

can be organized into graphs as shown in Figure 9(a)

and (b).

To evaluate the status of all students, we evaluate

the fragment graph bottom-up. During this process,

for each encountered node, we evaluate the VBSQT

represented by it and transfer the result obtained to

its parents. For example, consider Figure 9(a) for stu-

dents majoring in Geography and Physics. In a bot-

tom-up fashion, we first evaluate the SQTs

represented by v

, v

, and v

. The results are then

partitioned according to their majors and sent to the

corresponding parent nodes. In the next step, the

SQTs represented by the nodes labelled with Geogra-

phy and Physics are evaluated to find all the students

eligible to graduate.

The above process can be improved by using the

constants appearing in a query to speed up the com-

putation. For example, if we are considering only the

students majoring in Geography, we need to access

only part of the graph (marked grey) shown in Figure

9(a) or Figure 9(b).

<GeographyRule title=“Degree Requirement

for 3-Year BSc (Geography)”

combining=”AND”>

< Xinclude href=”GraduationRule.xml” >

<ResidenceRule title = ”Residence Requirement”

combining = ”AND” >

< Xinclude href=”DegreeRule.xml” >

<MajorRule title=”Major”, display=”minimum 18 credit hours”

query=”... ” ...>

</MajorRule>

</ResidenceRule>

< Xinclude href=”GeneralRule.xml” >

<MajorRule> title=”Major Requirement”

combining=”AND”>

<MinMaxRule

display=”Minimum 30 credit hours, Maximum 48 credit hours”

query=”... “ ...>

</MinMaxRule>”

<ReqCoursesRule title=”Required Courses”

combining=”AND”>

<Course

display=”23.202 Intro Geography I” query=”...” ...> </Course>

<Course

display=”23.203 Intro Geography II” query=”...” ...> </Course>

<Course

display=”23.331 Advanced Geography” query=”...” ...> </Course>

</ReqCoursesRule>

<ChoiceRule title=”Choice”

display=”23.205 Atmos Sci or 23.206 Earth Sci”

combining=”OR”>

</ChoiceRule>

</MajorRule>

</GeographyRule>

Figure 8: Requirements document referencing fragment

using the XML Xinclude feature.

3-Year BSc

(Geography)

3-Year BSc

(Physics)

(a) Fragment graph derived from Figure 4

Figure 9: Fragment graphs.

3-Year BSc

(Geography)

3-Year BSc

(Physics)

(b) Fragment graph derived from Figures 4, 6, 7

3-Year BA

(English)

WEBIST 2006 - INTERNET TECHNOLOGY

204

4 CONCLUSION AND FUTURE

WORK

In this paper, we consider a document type that

includes requirements and where a user comprehends

these requirements as rules to be followed to achieve

a certain designation. As a result, we consider each

document a single compound rule that may be

assembled from many fragments. When such a docu-

ment (e.g. 3-Year BSc Geography) is evaluated in a

certain context (e.g. for a specific student) there will

be a value generated for it. For this type of document,

fragmented SQTs succinctly represent document

content, evaluation and query requirements; a simple

tree traversal is required to evaluate or display a doc-

ument.

Since some requirements may appear many times

in different documents, these documents can exhibit

a great deal of redundancy. We have introduced an

algorithm to fragment a collection of documents, and

described an efficient approach for document evalua-

tion where each fragment/VBSQT is evaluated just

once, but many results are made available (i.e. for a

set of students).

We have developed a prototype system that

assembles and displays requirements documents

from fragments and determines on request the gradu-

ation status for students on a) an individual basis or

b) a set-oriented approach for handling many stu-

dents at one time. The former is useful by an individ-

ual student to measure their own progress, and the

latter approach is useful in a university setting at say,

the end of term, when students should be graduating.

The prototype has been constructed using Java, a

SAX parser, and student history data stored in a

mySQL relational database. Requirements docu-

ments are stored as fragments (related via Xinclude)

that are independent XML documents. Various func-

tions such as Logical And, Logical Or, Minimum,

and Arithmetic Add required for the general synthe-

sized query tree have been implemented.

We are examining algorithms for document eval-

uation, query optimization, discovery of common

subtrees, and other processing models such as the

pipe and filter architecture (Albin 2003).

We are examining other situations to apply the

query tree approach. We have not used functions that

return data in XML format. Such functions can be

used to perform an include operation, in the same

way that we have used Xinclude. Also, if we allow

functions as found in (Abiteboul et al 2002, Abite-

boul et al 2003, Bonifati et al 2001) that invoke arbi-

trary Web Services returning XML then our model

allows, as a special case, Active XML documents.

We intend to examine other issues related to the

processing of these query-based documents includ-

ing concerns such as “Which major, given my cur-

rent status, permits me to graduate most quickly?” or

“What are the added requirements if I were to do a

double major in Geography and Physics, instead of a

single major in Geography”?

REFERENCES

Abiteboul, S., Benjelloun, O., Manolescu I., Milo, T.,

Weber, R., 2002, Active XML: peer-to-peer data and

web services integration (demo), in Proceedings of

VLDB.

Abiteboul, S., Bonifati, A., Cobena, G., Manolescu, I.,

Milo, T., 2003, Dynamic XML documents with distri-

bution and replication, in SIGMOD 2003, June 9-12,

2003, San Diego, CA, USA.

Albin, S. T., The art of software architecture: design meth-

ods and techniques, Wiley Publishing, 2003, 1

edi-

tion, ISBN 0471228869.

Bonifati, A., Ceri, S. and Paraboschi, S., 2001, Active rules

for XML: A new paradigm for E-services, in VLDB

Journal, 10, 39-47.

Business Rules Group, Defining business rules: What are

they really?, 3rd. edition, July 2000, http://www.busi-

nessrulesgroup.org.

McFadyen, R., Chen, Y., Chan, F-Y., 2005, XML-based

evaluation of synthesized queries, in 1st International

Conference on Web Information Systems and Technolo-

gies (WEBIST 2005), 24-31, Miami, USA.

Quint, V., Vatton, I., 2004, Techniques for authoring com-

plex XML documents, in ACM Symposium on Docu-

ment Engineering (DOCENG 2004), Milwaukee,

Wisconsin, USA.

Ross, R. G., 1997, The business rule book, Business Rule

Solutions, Houston, 2

edition.

Salminen, A., Tompa, F. W., 2001, Requirements for XML

document database systems, in ACM Symposium on

Document Engineering (DOCENG 2001), Atlanta,

Georgia, USA.

XML.org, http://www.w3.org/TR/xinclude/, retrieved June

8, 2005

RULE-BASED QUERY TREE EVALUATION OVER FRAGMENTED XML DOCUMENTS

205