FEDERATED MEDIATORS FOR QUERY COMPOSITE ANSWERS

Dong Cheng and Nacer Boudjlida

LORIA-University Henri Poincar

e Nancy 1

BP 239, 54506 Vandoeuvre L

es Nancy CEDEX(F)

Keywords:

Composite answer, Description Logics, Mediator, complementary.

Abstract:

The capture, the structuring and the exploitation of the expertise or the capabilities of an “object” (like a

business partner, an employee, a software component, a Web site, etc.) are crucial problems in various appli-

cations, like cooperative and distributed applications or e-business and e-commerce applications. The work

we describe in this paper concerns the advertising of the capabilities or the know-how of an object. The

capabilities are structured and organized in order to be used when searching for objects that satisfy a given

objective or that meet a given need. One of the originality of our proposal is in the nature of the answers the

intended system can return. Indeed, the answers are not Yes/No answers but they may be cooperative answers

in that sense that when no single object meets the search criteria, the system attempts to ﬁnd out what a set of

“complementary” objects do satisfy the whole search criteria, every object in the resulting set satisfying part

of the criteria. In this approach, Description Logics (DL) is used as a knowledge representation formalism and

classiﬁcation techniques are used as search mechanisms. The determination of the “complementary objects”

is founded on the DL complement concept.

1 INTRODUCTION

In many situations and applications, one is faced

to the problem of discovering “entities” that satisfy

given requirements. On the other hand, most often,

information retrieval in a distributed context, like the

World Wide Web, usually lacks selectivity and pro-

vides large answer-sets due to the fact that the avail-

able information sources are not well structured nor

well classiﬁed, as attested by the tremendous work

about Semantic Web (W3C, 2003a). Moreover, most

of the systems usually return yes/no answers (i.e. ei-

ther they ﬁnd out answers to a given query or they

fail). The work we describe here concerns the pub-

lishing of the capabilities of given entities (i.e. what

functionalities a given entity is offering, what exper-

tise it has and so on). The capabilities are organized

and structured in order to be exploited when search-

ing for entities that satisfy an objective or that meet

given needs (searching for Web services that offer

given functions, searching for a component in the

context of component-based design and component-

based programming, searching for a business part-

ner with a given expertise, looking for an employee

whose records and expertise satisfy a given work po-

sition proﬁle are application examples of our work).

It is clear that these needs require the capture and the

organization of the entities capabilities together with

the classiﬁcation of the entities and their capabilities.

In this work, we adopted an object-oriented knowl-

edge representation using description logics (DL-org,

2003). From the system architecture point of view, we

opted for a model based on distributed and coopera-

tive mediators (or traders). A signiﬁcant originality

of our approach resides in the type of answers we aim

at providing. Indeed, when no single entity satisﬁes

the search criteria, the systems attempts to determine

a set of complementary entities who, when grouped

together, satisfy the criteria.

The presentation is structured as follows. In

section 2, we expose some motivations together

with the architecture of the target system viewed as

a Knowledge-Based Management System (KBMS).

Section 3 brieﬂy introduces elements of description

logics and shows how description logics is used for

entities capabilities management and retrieval. Con-

cluding remarks are in section 4.

170

Cheng D. and Boudjlida N. (2004).

FEDERATED MEDIATORS FOR QUERY COMPOSITE ANSWERS.

In Proceedings of the Sixth International Conference on Enterprise Information Systems, pages 170-175

DOI: 10.5220/0002638601700175

 SciTePress

2 MOTIVATION AND

ARCHITECURE

The dynamic discovery of services or capabilities

an “entity” offers, has different application domains.

Component-based programming, electronic business

(e-business) and even enterprise knowledge manage-

ment (Nonaka and Takeuchi, 1995) are among the ap-

plication domains in which there is a need for the dis-

covery of services or capabilities an “entity” offers.

The only syntactic description of an entity’s capability

(like the signature of a software component’s service)

is not satisfactory when using that description for an-

swering a request: an additional semantic description

is required. Moreover, the elicitation of possible re-

lationships among services may contribute to ﬁnd out

“the best” service or the “the best complementary”

services that satisfy a search query.

In e-business, this approach can be applied in the

constitution of business alliances or when looking for

business partners. For example, when trying to con-

stitute a business alliance, the system we aim to de-

velop services that may help in retrieving the most ap-

propriate candidate partners. Furthermore, the work

depicted hereafter may also serve for semantic-based

discovery of Web services (W3C, 2003c) and it is

in line with current activities on semantic and ontol-

ogy for the Web (uddi.org, 2000; Dyck, 2002; W3C,

2003b; W3C, 2003a).

The purpose of this work requires formalizing and

structuring the knowledge that concern the “entities”

in a given application domain. We opted for Descrip-

tion Logics (DL) (DL-org, 2003; Horrocks, 2002b) as

a knowledge representation formalism: it has the no-

table merit that a single mechanism (the classiﬁcation

mechanism) serves at the same time to build and to

query extendible domain descriptions.

As a matter of comparison, in (Han et al., 1999),

“entities” are design fragments that are described

thanks to keywords, the relationships between the

fragments are constituted by metrics that measures

the similarity between fragments. In (Borgida and

Devanhu, 1999), entities are object-oriented software

components and description logics is used, notably, to

describe their intended semantics together with possi-

ble constraints involving objects methods.

In (Bouchikhi and Boudjlida, 1998), “entities” are

software objects and the capabilities of a software ob-

ject are speciﬁed giving a syntactic part (signatures

of the methods or operations the object offers) and a

semantic part expressed as logical expressions (a pre-

condition and a post-condition are associated with ev-

ery object’s method). The syntactic conformance of a

method to a query is trivial and the semantic confor-

mance uses theorem proving techniques.

One should notice that description logics has been

used in the database domain (Borgida, 1995; Boud-

jlida, 1995)

, not only for querying but also for

database schema design and integration (Calvanese

et al., 1998), for reasoning about queries (query

containment, query reﬁnement, query optimisation,

. . . ) (Beeri et al., 1997). From our concern, descrip-

tion logics is used for query purposes with the special

objective to produce more than “Yes or No” results.

From the system architecture point of view, we

choose a trader (also called mediator) based architec-

ture (OMG, 1992) very similar to the notion of dis-

covery agency in the Web services architecture. In

this architecture, an “entity”, called exporter, pub-

lishes (tells) its capabilities at one or more mediators

sites (see ﬁgure 1). Entities, called importers, send re-

quests (queries) to the mediator asking for exporters

ﬁtted with a given set of capabilities.

Figure 1: The Mediator-based Architecture

The mediator explores its knowledge base to try

to satisfy the query. The capability search process is

founded on the exported capabilities and on relation-

ships between them, these relationships being trans-

parently established by the mediator. When some ex-

porters known from the mediator satisfy the query, the

references of those exporters are sent back to the im-

porter in L

answer

. Nevertheless, satisfying the query

falls into different cases (Boudjlida, 2002) (see ﬁgure

2):

• Case1: There exist exporters that exactly satisfy the

query;

• Case2: There exist exporters that fully satisfy the

query, but their capabilities are wider than those re-

quested;

• Case3: There exists no single exporters that fully

satisfy the query, but when “combining” or com-

posing capabilities from different exporter, one can

fully satisfy the query;

• Case4: Neither a single exporter nor multiple ex-

porters satisfy the query, but there exist some ex-

porters that partly satisfy the query;

• Case5: No single exporter nor several exporters

fully or partly satisfy the query.

See also (DL-org, 2003) for the proceedings of the vari-

ous “Knowledge Representation meets DataBase” (KRDB)

Workshops.

FEDERATED MEDIATORS FOR QUERY COMPOSITE ANSWERS

171

Figure 2: The cases of satisfaction

One should notice that cases 4 and 5 would con-

duct to a failure of the query when only one mediator

is implied. But, if we assume a grouping of media-

tors (into a federation of mediators), these cases are

typical cases where cooperation among the mediators

is required. In the case 5, the whole query is trans-

mitted for evaluation to other mediators whereas in

the case 4, we need to determine “what is missing” to

the individuals to satisfy Q, that means to determine

what part of the query is not satisﬁed by the found in-

dividuals. This part as well as the original query are

transmitted then to a mediator of the federation. Con-

ceptually, we can see the query as being addressed to

“the union” of by the federated mediators’ knowledge

bases. Concretely, this union is explored from “near

to near” within the federation, that means from a me-

diator to an other.

Let us now elaborate more on the conceptual

framework and the underlying formal foundations.

Given a query Q expressed as a sentence of a

query language L

Query

, a simple result (noted

Answer

(Q)) is generally returned in most of the

systems and usually L

Answer

(Q)

={Yes, No, Un-

known}, the meaning of “Yes” is that one or sev-

eral entities satisfy the L

Query

and “No” is the oppo-

site. In this work, we try to offer a L

Answer

(Q) that

may not be a single entity satisfying L

Query

, but that

may possibly be a collection of entities where “the

union” of the members of the collection of entities

satisﬁes the search criteria of L

Query

. This collection

can be found in a knowledge base or in a federation

of knowledge bases (that means a set of knowledge

bases).

In this work, we adopted a Description Logic lan-

guage (DL) (DL-org, 2003), that were intensively de-

veloped and studied in the ﬁeld of Knowledge Repre-

sentation. So it is not surprising that they are partic-

ularly adapted for representing the semantics of real

world situations including data semantics (Calvanese

et al., 1998; Borgida, 1995). The query language is

Figure 3: A model of composite answer

deﬁned by L

Query

={ DLs formulaes}, and L

Answer

is deﬁned by two components the L

Satisfaction

and

Complement

. L

Satisfaction

describes a satisfaction,

that is a single entity or a set of entities satisfying a

query Q. If Q is completely satisﬁed, its complement

(noted L

Complement

(Q)) will be empty. In the con-

trary, the system will try to determine a complement

for this answer. We deﬁne a function Comp(−, −)

to calculate the complement of an answer to a query:

Complement

(Q) = Comp(L

Satisfaction

(Q), Q).

Intuitively this complement designates “the missing

part” to an entity in order for that entity to satisfy the

query.

The ultimate goal of this work is the design and the

development of a set of services to export, import and

mediate, as well as the study and the development of

alternative strategies and mechanisms for mediators

cooperation. The coming sections detail the adopted

approach and its foundations, beginning with a short

introduction to description logics.

3 COMPLEMENT AND

COMPOSITE ANSWER

The determination of unsatisﬁed part of a query is

founded on the concept of complement (Schmidt-

Schauss and Smolka, 1991) in DLs and the test of

the relation of subsumption between the concepts of

the query and those in the mediator’s base. The algo-

rithm that we propose to test if this relation exists or

not, presents the originality, in the case where the re-

lation is not true, to identify the concepts of the query

that are the reason of the non existence of the relation

(concretely, it is about the concepts of the query that

are not subsumed by the concepts in the mediator’s

base): these concepts describe “the part that is miss-

ing” to the individuals who partly satisfy the query.

Let us now describe the proposal in a more formal

way, beginning by deﬁning the notion of complement

to arrive progressively to the procedure of its calcula-

tion.

3.1 An introduction to DLs

DLs (DL-org, 2003; Horrocks, 2002b; Napoli, 1997)

is a family of knowledge representation languages

ICEIS 2004 - SOFTWARE AGENTS AND INTERNET COMPUTING

172

where a description of a world is built using concepts,

roles and individuals. The concepts model classes

(sets of concepts, TBox)of individuals (sets of indi-

viduals, ABox) and they correspond to generic enti-

ties in an application domain. An individual is an in-

stance of a concept. Roles model binary relationships

among the individual classes. A concept is speciﬁed

thanks to a structured description that is built giving

constructors that introduce the roles associated with

the concept and possible restrictions associated with

some roles. Usually, the restrictions constrain the

range of the binary relationship that is deﬁned by a

role and the role’s cardinality.

PERSON

v TOP

SET

v TOP

MAN

v PERSON

WOMAN

v (and PERSON

(not MAN))

member

v toprole

chef

v member

TEAM

= (and SET

(all member PERSON)

(atleast 2 member))

SMALL-TEAM

= (and TEAM

(atmost 5 member))

MODERN-TEAM

= (and TEAM

(atmost 4 member)

(atleast 1 chef)

(all chef WOMAN))

Figure 4: An exemple of LDs TBox

Concepts are of two types, primitive and deﬁned

concepts. Primitive concepts may be considered as

atoms that may serve to build new concepts (the de-

ﬁned concepts). Similarly, roles may be primitive

roles as well as deﬁned roles. In the ﬁgure 4, PER-

SON and SET are primitive concepts: they are in-

troduced using the symbol

v and they are linked to

a “TOP” concept (>)

; TEAM SMALL-TEAM and

MODERN-TEAM are deﬁned concepts (they are in-

troduced using the symbol

=. The and constructor

enables deﬁning concepts as a conjunction of con-

cepts: these concepts are the immediate ascendants

of the deﬁned one. The all constructor constrains a

role’s range and the at-least and at-most constructors

enable specifying the role’s cardinals. Finally, the not

constructor only applies to primitive concept.

Subsumption is the fundamental relationship that

may hold among described concepts. Intuitively, a

concept C subsumes a concept D, if the set of indi-

Intuitively the TOP concept is the “most general one”

and it contains all the individuals while the BOTTOM con-

cept (⊥) is the most speciﬁc one and is empty.

viduals represented by C contains the set of individu-

als represented by D. More formally, C subsumes D

and it is denoted as D v C (or D is subsumed by C)

if and only if D

for every possible interpretation I.

C is called the subsuming concept and D is the sub-

sumed. For example in ﬁgure 4, PERSON subsumes

MAN, SMALL-TEAM is subsumed by TEAM.

The subsumption relationship organizes the con-

cepts into a hierarchy of concepts. The classiﬁca-

tion process aims at determining the position of a new

concept in a given hierarchy. In this framework, a

query is represented as a concept Q to be classiﬁed

in a given hierarchy. The result of the query is the

set of instances of the concepts that are subsumed by

Q. The classiﬁcation is a process that enables discov-

ering whether a subsumption relationship holds be-

tween a concept X, and those present in a hierarchy

H. The classiﬁcation process is decomposed into 3

steps:

• Retrieve the most speciﬁc subsuming concepts of

X (denoted MSSC(X));

• Retrieve the most general concepts subsumed by X

(denoted MGSC(X));

• (possibly) Update the links between X and its

MSSC and its MGSC.

The result of MSSC and MGSC can determine the

cases of satisfaction (see ﬁgure 5).

In DLs a classic deﬁnition of the complement[3]is:

given two descriptions A and B in ALCN R

A v B, the complementary of A relatively to B,

noted Comp(A, B), is deﬁned as the description C

is most general as A ≡ B u C. Comp(A, B) cor-

responds to the minimal additional knowledge that

is missing to a process of B to be an instance of

A. For example, Comp(SMALL-TEAM, TEAM) =

(at-most 5 member). The complementary of a con-

cept relatively to another concept and smallest sub-

suming common of two concepts. This deﬁnition

of the complement requires (B u C) ≡ A, where

(BuC) ≡ A ⇔ ((BuC) v A)∧((BuC) w A). The

determination of a complement is based on the sub-

sumption relationship as explained in the next section

where the test of the existence of the subsumption re-

lationship is based on a Normalize-Compare process.

3.2 Determination of the

Complement Concept

The aim of a normalization process is to put deﬁned

concepts to be compared, let say A and B, under a

conjunctive form: A = ( and A

, A

, . . . , A

)

and B = ( and B

, B

, . . . , B

). Once

ALCN R is a family of representation langage of DLs

FEDERATED MEDIATORS FOR QUERY COMPOSITE ANSWERS

173

normalized, two concepts can be easily com-

pared to check wither the subsumption relation-

ship holds between pairs of them or not: giving

A = ( and A

, A

, . . . , A

) and B = ( and

, B

, . . . , B

), the test “does the concept A

subsume the concept B?” returns “true”, if and only

if ∀A

(i ∈ 1,. . . ,n) ∃ B

(j ∈ 1,. . . ,m) such that

v A

The implementation of this process uses an array

of Boolean (called “Table Of Test” further) to record

the result of the subsumption relationship evaluation

(see ﬁgure 5). In that ﬁgure, C

, C

, · · · , C

de-

note the query concept under its normal form and

, D

, . . . , D

denotes the concepts “known

from” the mediators, i.e. every D

has to be

viewed under its normal form D

, D

, · · · , D

Then T able Of T est[D

, C

] = true means that

v C

. When the value returned by the func-

tion Subsumes(C, D

) is “false” (i.e. the con-

cept D

does not fully satisfy the concept C.),

therefore we need to determine a possible com-

plement of D

relatively to C. Using the Ta-

ble Of Test it is easy to get the complement of the

concept D

relatively to the concept C. For exam-

ple, “MODERN-TEAM subsumes SMALL-TEAM”

is false, “(atmost 4 member), (atleast 1 chef) and

(all chef WOMAN)” is false, then Comp(MODERN-

TEAM,SMALL-TEAM)=(and (atmost 4 member)

(atleast 1 chef) (all chef WOMAN)). Then us-

ing Table Of Test the deﬁnition of Comp() is:

Comp(C, D) =

k=1

, ∀k|T ableOf T est[k] =

false

. . . C

False False . . . True

False True . . . True

. . . . . . . . . . . . . . .

False False . . . False

ORoD False True . . . True

ORoS ANDoS

True False

Figure 5: Three parameters of “Table Of Test”

That means that the complement is given by the

conjunction of all the atomic concepts for which the

corresponding values in “Table Of Test” are “false”.

The composition of the truth values will permit us

to determine the cases of satisfaction. Let’s consider

a table ORoD[1..n] as ORoD[i] =

j=1

T [D

, C

ORoD[i] = true means that the concept C

satisﬁed by at least a D

. If the conjunction of

In the following, we will use T[x] instead of Table-

OfTest[x]

MSSC MGSC ORoS ANDoS CASE

X X True True 1

X ⊥ True True 2

> ⊥ True True 3

> ⊥ True False 4

> ⊥ False False 5

Figure 6: The analyse of the satisfaction case

the values of ORoD, noted AN DoS, is true (i.e.

i=1

ORoD[i] = T rue), it means that all the C

s are

satisﬁed and therefore the query. When AN DoS is

false, the logical disjunction of the values of ORoD,

noted ORoS, enables to determine a possible partial

satisfaction. Indeed if ORoS =

i=1

ORoD[i] =

T rue, it means that there exist some C

that are sat-

isﬁed. If both ORoS and AN DoS are false then no

atomic concept D

(j ∈ 1..m) satisﬁes a C

The ﬁgure 6

summarizes this discussion (in this

ﬁgure, the numbers in the CASE column refers to the

satisfaction cases listed in section 2). By this analyse

with the classiﬁcation algorithm (Boudjlida, 2002),

we can get a complete method to determine the sat-

isfaction cases that were listed in section 2.

3.3 Status of The Implementation

Based on these algorithms, an experimental platform

has been developed in Java. Some services, like

testing the subsomption relationship, determining the

complement of a concept and computing a compos-

ite answer, have been implemented on this platform.

All these services can be accessed through a Web

Server in a “classical” client/server architecture over

the Web. The services accept the DL concepts writ-

ten DAML+OIL (Horrocks, 2002a), an ontology lan-

guage in XML.

The DL concepts are encoded in XML and transmit-

ted to the Web Server who in turn transmit them to the

appropriate mediator service. Then a normalization

class transforms them into Java objects and an exper-

imental knowledge base, described in XML, is loaded

and normalized into a TBox object when the service is

started. All the algorithms of the mediator’s services

are implemented in the TBox class. The services’ out-

puts are also encoded in XML. Furhermore, XML is

also used to exchange information and concepts be-

tween the mediators when mediators’ copperation is

needed. For example, in the service that computes a

composite answer for a query concept, if the query

concept is not satisﬁed in the local knowledge base,

the complement concept is sent to next mediator as

In this table, X is a concept, > is the concept TOP and

⊥ is the concept BOTTOM.

ICEIS 2004 - SOFTWARE AGENTS AND INTERNET COMPUTING

174

an XML focument.

In the current status of the implementation, a medi-

ator “discovers” an other mediator using a static medi-

ator address list. More complex and dynamic discov-

ery techniques will be supported in the coming ver-

sions. Moreover, we deliberately ignored the search

of the actual individuals (ABox) that satisfy a query,

i.e. in the current work, we only consider TBoxes.

4 CONCLUSION

In this paper, we presented a method and an algo-

rithm for testing the subsumption relationship, deter-

mining concept complements and ﬁnding composite

answers. Some service of subsumption test and de-

termine Complement has been implemented in Java,

that is based on a normalization-comparison algo-

rithm and we can access these services by Web. One

of the originality of this work is in the type of query

answering we provide and also in the way we used

and implemented the complement concept. Indeed, to

the best of our knowledge, using the complement con-

cept for query answers composition does not ﬁgure

in the literature we had in hands nor in systems that

implement DLs, like RACER (Haaslev and Moller,

2003) one of the most representative DL system.

Future work may consider the complexity of the

algorithm, and the cooperation of mediators in het-

erogeneous environments, i.e. environments where

mediators’ knowledge bases are described in differ-

ent languages.

REFERENCES

Beeri, C., Levy, A., and Rousset, M.-C. (1997). Rewriting

Queries Using Views in Description Logics. In ACM

Symposium on Principles Of Database Systems, pages

99–108, Tucson, Arizona.

Borgida, A. (1995). Description Logics in Data Manage-

ment. IEEE Transactions on Knowledge and Data

Engineering, 7(5):671–682.

Borgida, A. and Devanhu, P. (1999). Adding more ”DL”

to IDL: Towards more Knowledgeable Component In-

teroperability. In 21rst International Conference on

Software Engineering, ICSE’99, pages 378–387, Los

Angeles, CA. ACM Press.

Bouchikhi, M. and Boudjlida, N. (1998). Using Larch to

Specify the Behavior of Objects in Open Distributed

Environments. In Proceedings of the 1998 Maghre-

bian Conference on Software Engineering and Artiﬁ-

cial Intelligence, pages 275–287, Tunis, Tunisia. 98-

R-300.

Boudjlida, N. (1995). Knowledge in Interoperable and

Evolutionary Systems. In Dreschler-Fischer, L. and

Pribbenow, S., editors, KRDB’95, Workshop on “Rea-

soning about Structured Objets: Knowledge Repre-

sentation Meets Databases”, pages 25–26, Bielefeld,

Germany. (Position Paper).

Boudjlida, N. (2002). A Mediator-Based Architecture for

Capability Management. In Hamza, M., editor, Pro-

ceedings of the 6th International Conference on Soft-

ware Engineering and Applications, SEA 2002, pages

45–50, MIT, Cambridge, MA.

Calvanese, D., de Giacomo, D., Lenzerini, M., Nardi, D.,

and Rosati, R. (1998). Information Integration: Co-

ceptual Modeling and Reasoning Support. In 6th In-

ternational Conference on Cooperative Information

Systems, CoopIS’98, pages 280–291.

DL-org (2003). Description logics. http//dl.kr.org/.

Dyck, T. (2002). Uddi 2.0 provides ties that bind.

(http://www.eweek.com/).

Haaslev, V. and Moller, R. (2003). Racer: Renamed

abox and concept expression reasoner. http://www.fh-

wedel.de/mo/racer/index.html.

Han, T.-D., Purao, S., and Storey, V. (1999). A Methodol-

ogy for Building a Repository of Object-Oriented De-

sign Fragments. In 18th International Conference on

Conceptual Modelling, ER’99, pages 203–217, Paris.

Spriger Verlag. LNCS 1728.

Horrocks, I. (2002a). DAML+OIL: a description logic for

the semantic web. IEEE Data Engineering Bulletin,

25(1):4–9.

Horrocks, I. (2002b). Description Logic: Axioms and

Rules. Talk given at Dagstuhl ”Rule Markup Tech-

nique” Workshop. http://www.cs.man.ac.uk/ hor-

rocks/Slides/.

Napoli, A. (1997). Une introduction aux logiques de de-

scription. Technical Report RR No 3314, INRIA-

LORIA, Nancy.

Nonaka, I. and Takeuchi, H. (1995). The Knowledge Creat-

ing Company; How Japanese Companies Create the

Dynamics of Innovation. Oxford University Press.

OMG (1992). The Object Model Architecture Guide. Tech-

nical Report 91.11.1, Revision 2.0, Object Manage-

ment Group.

Schmidt-Schauss, M. and Smolka, G. (1991). Attribute

Concepts Description with Complements. Artiﬁcial

Intelligence Journal, 48(1):1–26.

uddi.org (2000). UDDI: Universal Description, Dis-

covery and Integration. Technical White Paper.

(http://uddi.org).

W3C (2003a). Semantic Web. http://www.w3.org/2001/sw.

W3C (2003b). Web Ontology.

http://www.w3.org/2001/sw/WebOnt.

W3C (2003c). Web Services. http://www.w3.org/2002/ws.

FEDERATED MEDIATORS FOR QUERY COMPOSITE ANSWERS

175