resource space view tour mechanism - knowledge grid

Technical Report of Knowledge Grid Research Center, KGRC-2007-04, July, 2007, www.knowledgegrid.net/TR

Resource Space View Tour Mechanism

Jin Liu1, 2, Xiang Li1 and Liang Feng1

Key Lab of Intelligent Information Processing, Institute of Computing Technology, Chinese Academy of Sciences, Beijing, China1

State Key Lab of Software Engineering, Wuhan University, China2 {jliu, xiangli, feng_liang}@kg.ict.ac.cn

Abstract

The Resource Space Model is a new semantic data model for managing various resources.

Based on the model, this paper proposes a view mechanism for finding and reusing legacy

Resource Spaces according to users’ idiosyncratic interests. It establishes flexible reorganizing

mechanism of legacy Resource Spaces to reduce redundant work. As a viewpoint organization

strategy, the resource subject tree is proposed as an effective way to discover desirable Resource

Space views. A case study is presented to demonstrate the proposed approach. Key Words: Data model, Resource Space Model, Resource Space View, Viewpoint

1. Introduction

Zhuge’s Resource Space Model RSM is a novel semantic data model based on orthogonal

classification semantics, a set of normal forms and operations together with the normalization

and integrity theories to efficiently locate and operate versatile web resources with a universal

resource view [11, 12, 13, 14, 15, 16]. Recently, it develops towards a decentralized RSM by

incorporating Peer-to-Peer network and probability [17].

Legacy Resource Spaces contain rich domain knowledge and well-organized resources. To

reuse legacy Resource Spaces with respect to specific interests, this paper focuses on the

following two issues:

(1) Establishing a flexible reorganizing mechanism of legacy Resource Spaces in the

interconnection environment.

(2) Providing an effective way to discover desirable reorganized Resource Spaces.

This paper proposes the Resource Space view to solve the first issue. The view mechanism

includes the view construction and the resource fragmentation. The resource fragmentation is a


distributed application pattern set that guides the process of the view construction. A view can

be regarded as the result of iteratively applying the fragmentation patterns. A Resource Space

view as a representational scheme provides a viewpoint on a resource collection together with a

resource accessing mechanism. For a Resource Spaces centric application, a view is the major

medium that reifies a viewpoint. A resource subject tree as a viewpoint organization strategy is

proposed to establish an explicit association between Resource Space views and resource

subjects. With the knowledge synergy of the electronic index, the semantic network and

ontology, the subject tree owns the traits of explicit semantics and intuitive navigation

concerning human cognition, which makes the rapid resource viewpoint tour possible [5, 6, 10].

2. The Category Hierarchy of Resource Spaces A Resource Space RS(A, C) uses independent axes A(A1, A2,…, An) and axis coordinates

Ai{1i

C , …ni

C } to represent the classification semantics for versatile resources. Users locate and

operate resources by specifying a coordinate on each axis, denoted as a resource category. RSM

may contain a category hierarchy for observing the same objects from different abstract levels.

A coordinate Ci on an axis A can be split into a lower level coordinate set {1i

C , …ni

C }. The

position of the coordinate Ci in the axis A can be replaced by {1xi

C , …xniC }, denoted as axis

transformation. The transformed axis is orthogonal with any other axis in the original abstract

level. The comparative table of notations and explanation is provided to facilitate explanation

(see Appendix A).

Theorem 1. Let RS(A, C) be a Resource Space, where A{A1, ... , An} is an orthogonal axis set, C

is an axis coordinate set ∪Ai{1i

C , …hi

C }, 1 ≤ i ≤ n. Ai {1i

C , … ixC , ...

miC } has a category

hierarchy CH with ixC as its top. For ∀ Ck ∈ CH, Ck can split into an immediate low level

coordinate set {1xi

C , …nixC }. Ai can be transformed into Ai' that contains {

1xiC , …

nixC } at the

top level. It holds that Ai' is orthogonal with any other axes at the top level.

Proof 1. Let Ai {1i

C , … ixC , ...

miC }, 1 ≤ x ≤ m. For ∀At ∈{ A1, ... Ai-1} ∪ { Ai+1, ... An}, At

={1t

C , …kt

C }, At ⊥ Ai. Let the height of CH be H and the transformed axis be Ai' that contains


{1xi

C , …nixC }, denoted as Ai{

1iC , … ix

C , ... mi

C } ⇔ Ai'{1i

C , … 1ix

C−

, 1ix

C , …nix

C , CL,

1ixC

+, …

miC }. CL is the category set ⊆ CH. CL ∩ {

1xiC , …

nixC }=∅.

For H = 1, ixC can be split into an immediately low level coordination set {

1xiC , …

hixC }.

Ai{1i

C , … ixC , ...

miC } can be transformed into Ai'{

1iC , …

1ixC

−,

1ixC , …

hix

C , 1ix

C+

,

1ixC

+, …

miC }. CL = ∅. Since At ⊥ Ai, At finely classifies Ai, and ix

C can be split into an

immediate low level coordination set {1ix

C , … h

ixC }, it holds that

1ixC /At, …,

hix

C /At.

Therefore, it holds that 1i

C / At, …1ix

C / At, … hixC / At, …

miC / At, i.e., Ai' /At. For At{

1iC , …

ixC , …

miC }, and

1tC / Ai{

1iC , … ix

C , … mi

C },… kt

C / Ai{1i

C , … ixC , …

miC }. As a

result, 1t

C / Ai′{1i

C ,… 1

ixC ,…

nixC ,… mi

C },… kt

C / Ai′{1i

C , … 1ix

C , … nix

C , … mi

C }

holds, i.e., Ai / Ai'.

Therefore:

Ai ⊥ Ai' when H = 1 (1).

For H = k, k ≥ 1, we suppose:

Ai ⊥ Ai' when H = k (2)

For H = k+1, let Ck be a category whose abstract level in CH is k. Ck can be split into a set of

immediate low level coordinates {1

ixC , …

nixC }. Ai can be transformed into Ai(Ck) that

contains Ck at the top level. According to (2), Ai(Ck) is orthogonal with any other axes in the top

level, i.e., Ai ⊥ Ai(Ck). Therefore, At {1t

C , … kt

C } / Ai(Ck){1i

C ,… Ck, CL′,… mi

C } and

Ai(Ck){ 1i

C , …, Ck, CL′, … mi

C } / At {1t

C , … kt

C } holds. For Ck and {1ix

C , … nix

C }, the

actual height of its category hierarchy is 1. According to (1), At{1t

C , … kt

C } /Ai′ {1i

C ,…

1ixC ,…

nixC ,CL ,… mi

C } and Ai'{1i

C , … 1ix

C , … nixC , CL ,…

miC } / At{

1tC , …

ktC }

holds. Therefore Ai ⊥ Ai′ when H = k+1.

To summarize, Ai′ is orthogonal with other axes in the original abstract level.

3. The Resource Space View


3.1 Definition Definition 1. A Resource Space view is an virtual RS derived from materialized Resource

Spaces recursively using the following operation rules:

Let U(A, C) and W(A, C) be two Resource Spaces,

(1) The projection of U on its category subset X(AX, CX), denoted as U[X], is a view, where ∀Ai

∈ X(AX), Ai ∈ U(A), ∀Cj ∈ AXi(CX), Cj ∈ U(Ai(C)), 1 ≤ i ≤ |X(AX)| ≤ |U(A)|, 1 ≤ j ≤ |AXi(C)| ≤

|Ai(C)|, | ( )|1

U Ai=∪ U(Ai) = U(C).

(2) The equal axis merge of U and W, (U EAM∩ W) is a view, where ∀Ai ∈ U(A), Ai ∈ W(A),

|U(A)| = |W(A)| = |U EAM∩ W|, | |1

iAp=∪ (U EAM∩ W)(Ai, Cp) = ( | |

1iA

q=∪ (U EAM∩ W)(Ai, Cq))

∪ ( | |1

iAr=∪ (U EAM∩ W)(Ai, Cr)) , 1 ≤ i ≤ |U(A)|.

(3) The union of U and W, (U ∪ W) is a view, where ∀Instance(U∪W)( 1iC |A1, 2 j

C |A2, ... knC |

An) = InstanceU( 1iC |A1, 2 j

C |A2, ... knC |An) ∪InstanceW( 1i

C |A1, 2 jC |A2, ...,

knC |An), 1 ≤ i ≤

|A1(C)|, 1 ≤ j ≤ |A2(C)|, … 1 ≤ k ≤ |An(C)|.

(4) The difference of U and W, (U − W) is a view, where ∀Instance(U∪V) ( 1iC |A1, 2 j

C |A2, ...

knC |An) = InstanceU( 1iC |A1, 2 j

C |A2, ... knC |An ) − InstanceW( 1iC |A1, 2 j

C |A2, ... knC |An).

Table 1. The Interpretation of Primitive View Type

View Type Interpretation V = U[X] (∀r∈dom(V)) (M_V(r) = (∃u ∈ dom(U)) (u[AX, CX] = r AND M_U(u) = true))

V = U ∩EMA W (∀r ∈ dom(V)) (M_V(r) = (∃u ∈ dom(U)) (∃w ∈ dom(W)) (u[U] = r OR w[W] = r)),

where ∀ Ai ∈ U(A), Ai ∈ V(A), |U(A)| = |V(A)| = |U ∩EMA V|, | |1

iAp=∪ (U ∩EMA V)(Ai,

Cp) = ( | |1

iAq=∪ (U ∩EMA V)(Ai, Cq)) ∪ ( | |

1iA

r=∪ (U ∩EMA V)(Ai, Cr)), 1 ≤ i ≤ |U(A)|.

V = U ∪ W (∀r∈dom(V)) (M_V(r) = (r[U] OR r[W])), where ∀Ai∈ U(A), Ai ∈ V(A), |U(A)| =

|V(A)| = |U ∪ V|, ∀ ixC ∈U(Ai), ix

C ∈ W(Ai).

V = U − W (∀r∈dom(V))(M_V(r)=(r[U] AND ¬r[W])), where ∀Ai ∈U(A), Ai ∈ V(A), |U(A)| =

|V(A)| = |U ∪ V|, ∀ ixC ∈ U(Ai), ix

C ∈ W(Ai).

Using the domain of a Resource Space V to explain its meanings (denoted as M_V), Table 1


illustrates the meanings of four primitive view types. Let dom be a function that associates Ai

with its domain, denoted as dom(Ai) = (1i

C , … ixC , ...

miC ), then dom(V)=dom(A1) × dom(A2)

× …× dom(An) denotes the domain of the Resource Space V. An n-tuple ( 1xC /A1, …

xnC /An)

can specify a resource set with the same classification semantics. Thus the meaning of a view V

may be characterized according to the meanings of the Resource Spaces that deduce V.

To represent more complex resources, we extend the metadata description of Resource

Space to 2-dimensional geospatial resources and propose a corresponding data model in

accordance with the OGC specification [8] (see Appendix B). A 2-dimensional geospatial

resource describes the spatial geometry in a plane coordinate system.

For a view V with geospatial attributes V(A, C, G), dom(V) = dom(A1) × dom(A2) × …×

dom(An) × G. A view V with geospatial attributes V(A, C, G) can be derived from materialized

Resource Spaces according to two operation rules:

(1) The clip of G with GEOclip, U[GEOclip] is a view, where GEOclip is a reference geometry that

clips G.

(2) The union of U(A, C, G) and W(A', C', G'), U ∪Geo W is a view, where U(A, C) and W(A', C')

have the same schema about the axis and category set.

The view meaning for geospatial resources is: (1) V = U[GEOclip]: (∀r∈dom(V))(M_V(t) =

(r[V(A, C)] = r[V(A, C)] AND r[V(G)] = r[U(Clip(G, GEOclip))])); (2) V = U ∪Geo W: (∀r ∈

dom(V))(M_V(t) = (r[V(A, C)] = r[V(A, C)] AND r[V(G)] = r[Union(r[U(G)]), r[W(G')])]) ).

The definition statement of a Resource Space view is a SQL-like operation: CREATE

RS_VIEW <view-name> AS <view-definition>. Taking academic conferences as an example, a

Resource Space view Conference_view about International Conferences is derived from

Resource Space Paper_Conference1 at site1 and Resource Space Paper_Conference2 at site2, as

shown in the following SQL expression.

CREATE Conference_view RS_VIEW (Region, Year, Organization, Conference_Capital, Subject) AT site3 AS ( (SELECT (Region, Year, Organization, Conference_Capital, Subject) FROM Paper_Conference1 AT site1)

UNION SELECT (Region, Year, Organization, Conference_Capital, Subject) FROM Paper_Conference2 AT site2)

WHERE Paper_Conference1.Region = Paper_Conference2.Region


AND Paper_Conference1.Year = Paper_Conference2.Year AND Paper_Conference1.Organization= Paper_Conference2. Organization AND Paper_Conference1.Conference_Capital = Paper_Conference2.Conference_Capital AND Paper_Conference1.Subject = Paper_Conference2.Subject AND Year>=2000 & Year <= 2006);

For the view definition tree of Conference_view, the root represents the view and the leaves

represent Resource Spaces that derive Conference_view (see Figure 1). Other nodes on this tree

are the transition views for view definition. Along the path through the tree, the operation

expression for the Resource Space view Conference_view can be translated into the operation

expression for Resource Spaces that derives Conference_view.

Figure 1. Operation Expression on the View Definition Tree

Comparing a Resource Space view with a relational data view [3, 9, 18], both of them are

derived from materialized schemata. The reasons for the view mechanism in each data model

are that: (1)a view enables users to focus on their applications and ignore irrelevant data or

resources, (2)a view enhances the logical data independence or the resource independence by

concealing most changes of the materialized data model from users, (3)a view is a secure

strategy that prohibits accessing data or resources outside the view.

The differences between them stem from the goal and mathematical foundations of their

respective data models. While the relational data model focuses on data management, RSM

emphasizes the management of versatile resources in the interconnection environment [4, 11]. It

is more natural for RSM to manage resources since the orthogonal classification semantics is


familiar to human’s cognitive habits [11]. Moreover, while the relational data model grounds on

functional dependency [4], RSM accentuates the semantic independence and the semantic

orthogonality [11].

3.2 The Resource Fragmentation Many Resource Space centric applications exist as Resource Space views. These resource

schemata are logically uniform but actually partitioned into several category fragments derived

from different Resource Spaces.

The resource fragmentation divides the global Resource Space GRS (a view) into several

logical fragments. These fragments, stemming from the logical Resource Space LgRS, are

allocated to distributed sites. The Global Resource Space GRS (X(C), Q, S) reifies the global

Resource Space schema RSSGlobal as the structure and semantic restriction on GRS.

Definition 2. The 6-tuple Global Resource Space Schema RSSGlobal (X(C), D, DOM, Ι, Q, S),is

denoted as RSSGlobal. RSSGlobal is the name of Global Resource Schema; X(C) is the category

sequence comprising RSSGlobal; D is the range of C values; DOM is the mapping from X(C) to D;

Ι is the completeness restriction on RSSGlobal; Q is the restriction on C; S is the distributed

structure of RSSGlobal.

For a valid resource fragmentation, the resource partition satisfies three intrinsic

restrictions: integrity, reconstruction and disjointness.

Integrity as a mandatory restriction constrains that a resource description should at least

belong to a resource fragment if it belongs to GRS. That is, if a GRS is decomposed into LgRS1,

LgRS2, ..., LgRSn, for any resource description item rd, ∀rd ∈ GRS → ∃LgRSi, rd ∈ LgRSi′S

(one or more LgRSi, 1 ≤ i ≤ n).

Reconstruction as a mandatory restriction constrains that all fragments should be able to

reconstruct GRS, FS(GRS)=LgRS ∧ FS－1 (LgRS)=GRS. That is, if a GRS is decomposed into

LgRS1, LgRS2, ..., LgRSn, GRS = LgRS1 ∪ LgRS2 ∪ ... ∪ LgRSn, where ∪ is a merge operation.

Disjointness as an optional restriction constrains that any two fragments should not be

overlapped with each other, which helps to reduce redundancy of resource descriptions. That is,

if a GRS is decomposed into LgRS1, LgRS2, … LgRSn, LgRSi ∩ LgRSj = ∅, for any i, j, 1 ≤ i, j ≤

n and i ≠ j.


The resource fragmentation as the primitive application patterns guides designers to

construct distributed Resource Space applications. A global view can be the result of iteratively

applying these patterns. For the primitive RSM, the category correlation is a critical factor to

form the following resource fragmentation patterns.

The mono-axis fragmentation partitions a global Resource Space GRS into several resource

fragments by dividing an axis of this GRS according to the fragmentation conditions.

Definition 3. The mono-axis fragmentation MOF on GRS (X(C), Q, S) is an operation to

partition GRS into several resource fragments RF1(Xk(C1), Q1, S1), RF2(Xk(C2), Q2, S2), …

RFm(Xk(Cm), Qm, Sm) according to a group of category sequences Xk(CS1), Xk(CS2), …, Xk(CSm)

and holds that

(1) ⎪X(C)⎪ = n;

(2) Xk(Ci) = Xk(CSi), Xk(C)is the kth category sequence of X(C);

(3) Q1 = Q2 =…= Qm = Q;

(4) Si = ∅;

for any i, j, 1 ≤ i, j ≤ m, i ≠ j, Xk(CSi) ⊆ Xk(C), Xk(CSi) ∩ Xk(CSj ) = ∅, ∪ Xk(CSi) = Xk(C),

denoted as GRS(MOF)<Xk(CS)> = {RF1, RF2, … RFn}, where Xk(CS) = {Xk(CS1), Xk(CS2), …

Xk(CSm)}.

Figure. 2 The Mono-axis Fragmentation GRS (MOF)<T>={RF1,RF2,RF3}

Taking Paper_Conference as a global Resource Space GRS, Figure 2 illustrates that MOF

partitions the axis Region of Paper_Conference into three sections: {North_Ame, South_Ame,

Asia}, {Europe, Africa} and {Oceania}.


Character 1. MOF satisfies the intrinsic restrictions: integrity, reconstruction and disjointness.

Proof 2. Since ∪Xk(CSi) = Xk(C) and for any resource description item rd: ∀rd ∈ GRS → ∃RFi,

rd ∈ RFi, MOF satisfies integrity. MOF satisfies reconstruction and disjointness because

∪Xk(CSi) = Xk(C) and Xk(CSi) ∩ Xk(CSj )=∅.

The multi-axes fragmentation partitions a GRS into several resource fragments by dividing

more than one axis of the GRS according to the fragmentation conditions.

Definition 4. The multi-axes fragmentation MUF on GRS (X(C), Q, S) is an operation to

partition GRS into several resource fragments RF1(Xk1(C1), Q1, S1), RF2(Xk1(C2), Q2, S2), …

RFm(Xkh(Cm), Qm, Sm) according to a group of category sequences Xk1(CS1), Xk2(CS2), …,

Xkh(CSm) and holds that

(1) ⎪X(C)⎪ = n;

(2) 1< kr ≤ kh ≤n, the kh axis is a divided axis in fragmentation;

(3) Xkr(Ci) = Xkr(CSi), Xk1(C)is the k1th category sequence of X(C);

(4) Q1 = Q2 = … = Qm = Q;

(5) Si = ∅;

for any i, j, 1≤ i, j ≤ m, i ≠ j, Xkr(CSi) ⊆ Xkr(C), Xkr(CSi) ∩ Xkr(CSj)=∅, ∪Xkr(CSi)= Xkr(C) and the

multi-axes fragmentation is denoted as GRS(MUF)<Xk1, … kh(CS)> = {RF1, RF2, … RFn}, where

Xk(CS) = {Xk1(CS1), Xk1(CS2), … Xkh(CSm)}.

Figure 3. The Multi-axis Fragmentation GRS(MUF)<T>={RF1, … RF9}

Taking Resource Space paper_conference as a global Resource Space GRS, Figure 3


illustrates that MUF partitions the axis Region and Year of Paper_Conference into nine sections:

({North_Ame, South_Ame, Asia}, {Europe, Africa}, {Oceania}) × ({2004}, {2005}, {2006}).

Character 2. MUF satisfies the intrinsic restrictions.

Proof 3. Since ∪Xkr(CSi) = Xkr(C), 1< kr ≤ kh ≤ n and for any resource description item rd: ∀rd

∈ GRS → ∃RFi, rd ∈ RFi, MUF satisfies integrity. MUF satisfies reconstruction and

disjointness because ∪Xkr(CSi) = Xkr(C), 1 < kr ≤ kh ≤ n, Xkr(CSi) ∩ Xkr(CSj) = ∅.

The induced-axis fragmentation partitions a GRS into several resource fragments by

simultaneously dividing several axes of this GRS according to the association relationship

between these axes and the axes of other reference Resource Spaces.

Definition 5. The induced-axis fragmentation INF on the GRS (X(C), Q, S) is an operation to

partition GRS into several resource fragments RF1(Xk(C1), Q1, S1), RF2(Xk(C2), Q2, S2), …

RFm(Xk(Cm), Qm, Sm) according to another GRST(X(C), Q, S). GRST(X(C′), Q′, S′) has been

divided into resource fragments RFT1(Xk(C1′), Q1′, S1′), RFT2(Xk(C2′), Q2′, S2′), … RFTm(Xk(Cm′),

Qm′, Sm′) according to GRST(MOF)<Xk(CS)>={RFT1, RFT2, … RFTm}. INF holds that

(1) ⎪X(C)⎪ = n;

(2) Xk(Ci) ⊆ Xk(Ci′), Xk(C)is the kth category sequence of X(C);

(3) Q1 = Q2 = … = Qm = Q;

(4) Si = ∅;

for any i, j, 1 ≤ i, j ≤ m, i ≠ j, Xk(Ci) ⊆ Xk(C), Xk(CSi′) ∩ Xk(CSj′) = ∅, ∪ Xk(CSi) = Xk(C),

∪Xk(CSi′) = Xk(C′), and the induced-axis fragmentation is denoted as GRS(INF)<T> = {RF1,

RF2, … RFn}, where T = {RFT1, RFT2, … RFTm}.

Character 3. INF satisfies the intrinsic restrictions.

Proof 4. Since ∪Xk(Ci)=Xk(C) and for any resource description item rd: ∀rd ∈ GRS → ∃RFi, rd

∈ RFi, INF satisfies integrity. INF satisfies reconstruction and disjointness because ∪Xk(Ci) =

Xk(C) and Xk(Ci′) ∩ Xk (Cj′ ) = ∅ → Xk(Ci) ∩ Xk(Cj ) = ∅.

Taking two Resource Space Paper_ Conference and Review as two global Resource Space

GRS, they own the same axis Region. The axis Region of Paper_Conference is divided into

three sections: {North_Ame, South_Ame, Asia}, {Europe, Africa} and {Oceania}. INF

partitions the axis Region of Review into three sections according to the partition of the axis

Region of Paper_Conference (see Figure 4).


Figure. 4 The Induced-axis Fragmentation GRS (INF)<T> = {RF1, RF2, RF3}

To support the view involving geospatial resources, resource fragmentations are extended

to the geospatial fragmentation. The geospatial fragmentation GF partitions a GRS that contains

geospatial attributes into several resource fragments by clipping geospatial objects into several

geometry sections within valid regions.

Definition 6. The geospatial fragmentation GF on the GRS(X(C), G, REG, Q, S) is an operation

to partition GRS into several resource fragments RF1(X(C1), G1, REG1, Q1, S1), RF2(X(C2), G2,

REG2, Q2, S2), … RFm(X(Cm), Gm, REGm, Qm, Sm) according to a 2-tuple sequence (<GEOclip_1,

REGclip_1>, <GEOclip_2, REGclip_2>, … <GEOclip_m, REGclip_m>), where GEOclip_h is a reference

geometry and REGclip_h is a valid spatial region, 1≤ h ≤ m. GF holds that

(1) X(C1) = … X(Cm) = X(C);

(2) Gh = Clip(G, GEOclip_h);

(3) REGh = REGclip_h;

(4) Validate(Gh, REGh) ≡ true;

(5) Union(Clip(G, GEOclip_1), Clip(G, GEOclip_2), … Clip(G, GEOclip_m)) = G;

(6) Union(REG1, REG2,… REGm) = REG;


(7) Q1 = … = Qm= Q;

(8) Si = ∅;

for any i, j, 1 ≤ i, j ≤ m, i ≠ j, Relate(REGi, REGj, OVERLAP) ≡ true, denoted as

GRS(GF)(<GEO, REG>) = {RF1, RF2, … RFn}, <GEO, REG> = {<GEO1, REG1>, <GEO2,

REG2>, … <GEOm, REGm>}.

Character 4. GF satisfies the intrinsic restrictions.

Proof 5. According to the restrictions (1), (5) and (6) in definition 6, for any resource

description item rd: ∀rd ∈ GRS → ∃RFi, rd ∈ RFi, therefore GF satisfies integrity and

reconstruction. GF also satisfies the restrictions of disjointness because Validate(Gh, REGh) ≡

true and Relate(REGi, REGj, TOUCH) ∨ Relate(REGi, REGj, DISJOINT) ≡ true.

A Case of GF is illustrated in section 5 (see Figure 7).

The mixed fragmentations are omitted for the sake of paper’s length limitation, which

include the mono-axis & geospatial fragmentation MOGF, the multi-axis & geospatial

fragmentation MUGF and the inductive & geospatial fragmentation INGF. These proposed

fragmentations enable partition of complex geospatial resources. It is also provable that MOGF,

MUGF and INGF satisfy the aforementioned intrinsic restrictions.

4. The Simple Resource Subject Organization

4.1 The Subject-centric Viewpoint Organization Based on the Resource Space view, the general process of reusing legacy Resource Spaces

involves that: (1)users put forward the resource requests with respect to their applications;

(2)users find the Resource Space viewpoints according to their desirable resource characters;

(3)users access resources via viewpoints. A Resource Space view as a representational scheme is

the major medium to reify a viewpoint on a resource collection.

The investigation on the resource viewpoint location indicates that the discrepancy

between human’s abstract mental world and the virtual resource world prevents users from

efficiently finding Resource Space views according to their idiosyncratic interests.

Our solution is to organize subjects for resource viewpoints into a resource subject tree

because the resource interests of Resource Space schemata could be represented with concept


subjects. It is inspired from the topic map based on the electronic index, the semantic network

and ontology [6]. As a structural knowledge paradigm for finding information in books, the

electronic index of libraries adopts the subject-based classification to group documents

according to their content subjects. The semantic network endows the subject tree with the

adequate knowledge expressive ability and the intuitive information understanding for human

beings. Ontology provides an explicit understanding of domain concepts [6].

With this knowledge synergy, a resource subject tree indicates its users where to go to find

their desirable resource viewpoints in the interconnection environment and achieves a

subject-centric viewpoint organization.

4.2 The Resource Subject Tree By attaching resource interest of Resource Space views to concept subjects and taking the tree

as the organizing structure, an explicit association between views and resource subjects can be

established. Accordingly, the subject tree as a resource viewpoint organization strategy is

proposed to describe the subject configuration.

Assumption 1. The backbone topology of the resource subject organization based on view is a

hierarchical tree structure. Each node on this tree is a concept subject to annotate views.

Definition 7. The resource subject organization is a backbone tree TP = (P, E, O), where P ={p1,

… pn} is a node set and E is a immediate linkage set on P. ∀pi ∈ P is a subject node and p1 is

the root node on the tree. For ∀(pi , pj) ∈ E, 1 ≤ i, j ≤ n, i ≠ j, pi is the parent node and pj is a

child node. Node pi can be represented as a 3-tuple pi(Si, SDi, Vi(<Vh, Dh>)), where Si is a subject

concept selected from a ontology set O; SDi is a subject description of Si; for h ≥ 0, Vi(<Vh, Dh>)

is the view-description sequence where Vh denotes a view and Dh denotes the view description.

The backbone tree can be further denoted as TP=(pi(Si, SDi, Vi(<Vh, Dh>)), E, O).

The backbone tree provides a reference for building a knowledge representation hierarchy

and enables navigation on this generalization hierarchy (see Figure 5). For ∀ (pi , pj) ∈ E, 1 ≤ i, j

≤ n, i ≠ j, the broader/narrower relationship is used to build the backbone hierarchy, where pi is

broader in subject meaning than pj. Most subject node Si in the tree is picked from the formal

ontology set O. SDi is the lingual subject description for Si. With the intuitive but formal subject

tree and the equivalent natural language, this bi-modal of the tree provides a unified way


catering to human perceptions as well as being machine-processable.

Figure 5. The Backbone Resource Subject Tree TP

A view with a category hierarchy, denoted as VH, can be decomposed into several

unwrapped views according to the folding categories and the abstract levels. These unwrapped

views can be organized into a view tree according to the abstract levels, denoted as TVH. The

subject nodes of the unwrapped views tend to be placed below the nodes of the wrapped view

on the tree hierarchy (see Figure 5). Taking TVH as a sub-tree with respect to the backbone tree

TP, the fusion of the abstract-refinement mechanism into the backbone resource tree enables the

intuitive zoom-in and zoom-out navigation over the same schema but across different abstract

levels.

We extend the resource subject tree to represent correlative relationships such as synonymy

and similarity beyond the broader / narrower relationship.

Assumption 2. The resource subject tree can be extended by adding the associative relationship

that complements to the broader/narrower relationship.

Definition 8. The extended resource subject tree can be denoted with a tree TP' = (pi (Si, SDi,

Vi(<Vh, Dh>)), E, O, Rel), where Rel is the relationship function Rel: E → S, S ≡ {general,

associative}.

The Simple Knowledge Organization System SKOS was chosen to implement the subject

tree [7] (see Figure 6.). As a scalable framework of content subjects for human and machines,

SKOS promotes the semantic interoperability in the interconnection environment.


Figure 6. The SKOS Graph of Extended Backbone Tree TP'

5. Scientific Research Platform We implement the proposed resource viewpoint tour mechanism in an e-Science platform for

scientific researchers. This platform is an ongoing research for promoting individual and

cooperative research among scientists in Computer Science CS and Computer Engineering CE.

The AKT CS Reference Ontology and the ACM Computing Classification System are chosen as

domain ontologies to facilitate constructing the resource subject tree [1, 2]. List 1 demonstrates

a SKOS segment containing the definition of a resource subject, a view, and the zoom-in &

zoom-out relationship between a wrapped view and an unwrapped view in the resource subject

tree.

List 1. A SKOS Segment of the Resource Subject Tree

/*Subject*/ <skos:Concept rdf:about="/treesubject/#subject1"> <skos:prefLabel>Conference</skos:prefLabel> <skos:scopeNote

rdf:resource="notes/topnode1.txt"/> <skos:narrower> <skos:Collection> <skos:member rdf:resource="


treesubject/#subject11"/> <skos:member rdf:resource=" treesubject/#subject12"/>

<skos:member rdf:resource="treesubject/#subject13"/> </skos:Collection> </skos:narrower> </skos:Concept> /*Attach a view and its description to subject*/ <rs:view rdf:about="rs/rsview12"> <skos:subject rdf:resource="

treesubject/#subject12"/> <rs:description rdf:resource="description/rsview1.pdf"/> </rs:view> /*Zoom-in and Zoom-out*/ <rs:view rdf:about=" rs/rsview12"> </rs:view> <rs:view rdf:about="rs/rsview121"> <skos:prefLabel>SubjectTerm12</skos:prefLabel> <rs:zoomout rdf:resource="rs/rsview12"/> </rs:view>

The resource fragmentation as the reference pattern facilitates constructing Resource Space

views. Derived from three legacy Resource Spaces Central_reviewer, North_reviewer and

East_reviewer, a Resource Space view Grid_reviewer concerns location information of

conference reviewers in China (see Figure 7).

Figure 7. Defining View According to the Resource Fragmentation GF

With a similar schema structure, these legacy Resource Spaces involve in the

administrative regions: North China r1, Central China r2 and East China r3 respectively. The

view Grid_reviewer provides users a unified viewpoint concerning reviewers within these

regions. According to the geospatial fragmentation GF, Grid_reviewer as the Global Resource

Space GRS (X(C), G, REG, Q, S) is partitioned into three detached logical resource fragments

from Central_reviewer, North_reviewer and East_reviewer. The definition statement of the

view Grid_reviewer is as follows.


CREATE Grid_reviewer RS_VIEW (Name_Capital, Region, Gender, Age, Subject, Union(C01, C02, C03), Envelope(Union(C01, C02, C03)) ) AT site4 AS (

( SELECT (Name_Capital, Region, Gender, Status, Subject, C01=Clip(valid_NorthChina,admin_boundary02)) FROM Central_reviewer AT site1)

MERGE ( SELECT (Name_Capital, Region, Gender, Status, Subject, C02=Clip(valid_CentralChina, admin_boundary02))

FROM North_ reviewer AT site2) MERGE

( SELECT (Name_Capital, Region, Gender, Status, Subject, C03=Clip(valid_EastChina, admin_boundary03)) FROM East_reviewer AT site3)

WHERE Central_reviewer. Name_Capital = North_ reviewer. Name_Capital AND North_ reviewer. Name_ Capital = East_reviewer. Name_Capital AND Central_ reviewer. Region = North_reviewer. Region AND North_ reviewer. Region =East_reviewer. Region AND Central_ reviewer.Gender = North_ reviewer.Gender AND North_ reviewer. Gender=East_reviewer.Gender AND Central_reviewer.Age=North_reviewer.Age AND North_reviewer .Age=East_reviewer.Age AND Central_ reviewer.Subject=North_reviewer.Subject AND North_reviewer. Subject = East_reviewer.Subject AND C01= North_reviewer. Clip(valid_region01,admin_boundary01) AND C02=North_ reviewerClip(valid_region02,admin_boundary02) AND C03 =East_reviewer Clip(valid_region03, admin_boundary 03));

For these Resource Spaces, the administrative boundary is usually an arbitrary closed

contour. The spatial operations clip the administration regions and unite them within the valid

region of the view rsview.

According to the geospatial model, an XML schema has been developed to implement the

Resource Space for geospatial resources. With the XML segment for a resource instance

description, List 2 demonstrates an XML segment containing the Resource Space axes, the axis

coordinates, the geospatial description and the region description. This resource instance records

a female reviewer from China in Knowledge Grid field.

List 2. An XML Segment of the Resource Schema for Geospatial Resources

/*definition resource*/ <xs:element name="resource"> <xs:complexType> <xs:sequence> <xs:element ref="resource-name"/> <xs:element ref="description"/> </xs:sequence> </xs:complexType> </xs:element> /*basic RSM, geospatial and region description*/ <xs:element name="description"> <xs:complexType> <xs:sequence> <xs:element ref="RSM"/> <xs:element ref="Geospatial"/> <xs:element ref="Region"/> </xs:sequence> </xs:complexType>

</xs:element> /*basic RSM*/ <xs:element name="RSM"> <xs:complexType> <xs:sequence> <xs:element ref="Name_Capital"/> <xs:element ref="Region"/> <xs:element ref="Gender"/> <xs:element ref="Age"/> <xs:element ref="Subject"/> </xs:sequence> </xs:complexType> </xs:element> /*spatial attribute*/ <xs:element name="Geospatial"> <xs:complexType> <xs:sequence> <xs:element ref="type"/>


<xs:element ref="coordinates"/> <xs:element ref="element_sequence"/> <xs:element ref="ordinate_sequence"/> </xs:sequence> </xs:complexType> </xs:element> /* Resource instance fragmentation*/ <RSM>

<Name_Capital>J</Name_Capital> <Region>China</Region> <Gender>female</Gender> <Age>31</Age> <Subject>Knowledge Grid</Subject>

</RSM> <Geospatial>

<type>1</type> <coordinates>1</coordinates> <element_sequence>

<position>1</position> <element_info>100</element_info> <number>1</number>

</element_sequence> <ordinate_sequence>

<x>36.7</x> <y>52.9</y>

</ordinate_sequence> </Geospatial>

Figure 8. The Interface Tools of the Resource Subject Tree and Viewpoints

The user tools for browsing the resource subject tree and Resource Space views have been

developed with Java language and MySql database (see Figure 8). At the bottom of Figure 8,

users can browse a resource view to analyze resources about Knowledge Grid research. The

resource subject tree organizing resource subjects enables users to focus on their desirable

resources and keep the orientation during their touring through the subject hierarchy.

The number of the maximum fan-out for a node is limited to 9. Since the number of all tree


nodes is 99, the height for the tree would be 4 if the resource subject tree is a full tree. In this

case, the actual height for the source subject tree is 10. To estimate the efficiency of the resource

subject tree in resource viewpoint location, 20 applications concerning 20 Resource Space

views are prepared to compare the unstructured browse and our tree browse with the same

subject vocabulary. All views have been attached to these subjects. In total 10 volunteers are

equally divided into two groups to browse the unstructured subjects and the subject tree

respectively. Each group has been assigned with the same 6 randomly selected applications.

Figure 9. The Comparison between the Tree Browsing and the Unstructured Browsing

To locate the desirable viewpoints for each application, Figure 9 illustrates the browse

steps of every participator per task, and the mean browse steps per task for these two different

approaches. The result indicates that, for the same task, the maximum individual difference is 39

steps and the meaning individual difference is 17 steps. The minimum group difference reaches

59 steps. The mean group difference per task is 85.2 steps.

6. Conclusion This paper proposes to use the Resource Space view and the resource subject tree to construct a

viewpoint tour mechanism for RSM. It can reduce redundant work and make Resource Spaces

applicable to a wider range. The approach facilitates manifold applications with minimal

customization and broadens the coverage of the Resource Space centric resource sharing.


Further, the resource viewpoint organization relieves the discrepancy between human cognitive

world and the resource world. So it makes users easier to locate resources than the unstructured

resource browsing, as illustrated in the study case. Moreover, the investigation on the resource

fragmentation, involving the basis of the distributed Resource Space application, has been

preliminarily illustrated in the e-Science platform.

Acknowledges

This work was supported by the National Basic Research and Development 973 Program

(2003CB317000), the International Cooperation Program of the Ministry of Science and

Technology of China (2006DFA11970), the Natural Science Foundation of Hubei

(2005ABA240 & 2005ABA123), and the National Science Foundation of China (60503047).

Appendix

A．Notations and Explanation

Table 2. The Comparative Table of Notations and Explanation

Notation Full Name Notation Full Name A(C) A axis with C its category set MOF The mono-axis fragmentation CH A category hierarchy with H as its height. RS(A, C) The axis set A & and category set C

C|A1 C is a category of the axis A1 M_V The domain of the Resource Space Vdom(V) The domain of a Resource Space (view) MUF The multi-axes fragmentation

FS The fragmentation operation RF The resource fragmentation GEOclip The clip of G according to GEOclip RS Resource Space

GF The geospatial fragmentation RSM The Resource Space Model GRS The global Resource Space RSSGlobal The global Resource Space schema INF The induced-axis fragmentation ∩EAM The equal axis merge operation

LgRS The logical Resource Space ∪Geo The union operation on geometries

B．The Data Model for 2D Spatial Geometry We extend the metadata description of a Resource Space for 2-dimensional geospatial resources

and propose a data model complying with the OGC specification [8]. This data model has been

applied to the e-Science platform (see List 2).

Geospatial (T, C, E, O) is a 4-tuple that describes a 2-dimensional geospatial geometry. T

indicates the geometry type that consists of several geometry elements; C is used to identify a

coordinate system; E as a varying length array indicates the primitive element sequence that


constitutes geometry; O as a varying length array records the coordinate-value pair sequence

that makes up geometry boundary. Each element ei of E is a 3-tuple ei (Pi, EIi, Ni), where Pi

records its initial position in O, EI indicates the geometry information, and Number explains the

number of contiguous sub-elements. Table 3 explains the value-semantics of the geometry type

and Table 4 explains the value-semantics of the geometry element.

Table 3. The Value and Semantics of Geometry Type Type Semantics Type Semantics 0 Unknown 4 Point collection

1 Point 5 Line string or arc string collection

2 Line or arc 6 Polygon collection

3 Polygon 7 Heterogeneous collection

Table 4. The Value and Semantics of Geometry Element

ET L/A E/I N Semantics ET L/A E/I N Semantics

0 0 0 0 Unknown 3 2 1/2 0 Simple polygon connected by arcs.

1 0 0 n Point cluster 4 1 1/2 0 Rectangle described by the lower-left point and the upper- right point.

2 1 0 n Line string 5 2 1/2 0 Circle described by three different points on the circumference.

2 2 0 n Arc string 6 0 1/2 n Compound string connected by straight line strings or arc strings.

3 1 1/2 0 Simple polygon connected by straight lines.

7 0 0 n Compound polygon connected by straight lines or circular arcs.

Note: ET: geometry element type; L/A: line/arc, 1 denotes line string and 2 denotes arc string; E/I: exterior/interior, 1 denotes exterior polygon and 2 denotes interior polygon; N: the number of contiguous subelements constituting the element; 0 denotes meanings is unknown.

Clip and Union are crucial spatial operations for implementing the viewpoint centric

geospatial resource fragmentation, owning to their ability to partition and unite geometries.

Clip-statement. Grammar: Clip (Geo1, Geo2, Tolerance) RETURN Geo. The Clip statement

returns a geometry object that is the topological intersection of two geometry objects.

Union-statement. Grammar: Union (Geo1, Geo2, Tolerance) RETURN Geo. The Union

statement returns a geometry object that is the topological union of two geometry objects.

References [1] ACM Computing Classification System, 1998 Version, valid in 2006.

http://www.acm.org/class/1998/

[2] AKT Reference Ontology, http://www.aktors.org/ publications/ontology/, 2002.


[3] J. A. Blakeley, P. A. Larson, and F. W. Tompa, Efficiently Updating Materialized Views,

International Conference on Management of Data (SIGMOD 1986), ACM Press, 15(2),

1986, 61-71.

[4] E. F. Codd, Extending the Database Relational Model to Capture More Meaning, ACM

Transactions on Database Systems (TODS), 4(4), 1979, 397-434.

[5] N. Milic-Frayling, R. Jones, K. Rodden, G. Smyth, A. F. Blackwell and R. Sommerer,

Smartback: Supporting Users in Back Navigation, 13th International World Wide Web

Conference (WWW2004), 2004, 63-71.

[6] N. Guarino, Formal Ontology and Information Systems, 1st International Conference of

Formal Ontology in Information Systems (FOIS’98), IOS Press, 1998, 3-15.

[7] A. Miles and D. Brickley, editors. SKOS Core Guide, 2nd W3C Public Working Draft.

World Wide Web Consortium, 2005. http://www.w3.org/

[8] G. Reynolds, editor. OpenGIS Geographic Objects Implementation Specification, 2005.

http://www.open-geospatial.org/standards/

[9] O. Shmueli and A. Itai, Maintenance of Views, International Conference on Management

of Data (SIGMOD 1984), ACM Press, 1984, 240-255.

[10] H. Weinreich, H. Obendorf, E. Herder and M. Mayer, Off the Beaten Tracks: Exploring

Three Aspects of Web Navigation, 15th International World Wide Web Conference

(WWW 2006), 2006, 133-142.

[11] H. Zhuge, The Knowledge Grid, World Scientific Publishing Co., Singapore, 2004.

[12] H. Zhuge, Resource Space Grid: Model, Method and Platform. Concurrency and

Computation: Practice and Experience, 16 (14) (2004) 1385-1413.

[13] H. Zhuge, Resource Space Model, Its Design Method and Applications. Journal of Systems

and Software, 72(1) (2004) 71-81.

[14] H. Zhuge, A Knowledge Grid Model and Platform for Global Knowledge Sharing, Expert

Systems with Applications, 22 (2002) 313-320.

[15] H. Zhuge, E. Yao, Y. Xing, and J. Liu, Extended Normal Form Theory of Resource Space

Model, Future Generation Computer Systems, 21 (1) (2005) 189-198.

[16] H. Zhuge, Fuzzy Resource Space Model and Platform. Journal of Systems and Software,

73(3) (2004) 389-396.


[17] H. Zhuge, The Web Resource Space Model, Springer, 2007.

[18] Y. Zhuge, H. Garcia-Molina, J. Hammer and J. Widom, View Maintenance in a

Warehousing Environment, International Conference on Management of Data (SIGMOD

1995), ACM Press, 24 (2)(1995) 316-327.


resource space view tour mechanism - knowledge grid

Documents