IMPLEMENTATION OF INTENTION-DRIVEN SEARCH

PROCESSES BY SPARQL QUERIES

Olivier Corby

INRIA Sophia Antipolis, BP 93, FR-06902 Sophia Antipolis, Cedex France

Catherine Faron-Zucker

I3S, Universit´e de Nice Sophia, CNRS, BP 121, FR-06903 Sophia Antipolis, Cedex France

Isabelle Mirbel

INRIA Sophia Antipolis and I3S, Universit´e de Nice Sophia, CNRS, France

Keywords:

Search Process Modeling, Process Guidance, Ontology, Semantic Web, SPARQL, RDF(S).

Abstract:

Capitalisation of search processes becomes a real challenge in many domains. By search process, we mean

a sequence of queries that enables a community member to ﬁnd comprehensive and accurate information

by composing results from different information sources. In this paper we propose an intentional model

based on semantic Web technologies and models and aiming both at the capitalization, reuse and sharing of

queries into a community and at the organization of queries into formalized search processes. It is intended to

support knowledge transfer on information searches between expert and novice members inside a community.

Intention-driven search processes are represented by RDF datasets and operationalized by rules represented by

SPARQL queries and applied in backward chaining by using the CORESE semantic engine.

1 INTRODUCTION

The study of research projects aiming at support-

ing activities of community members through a col-

lective semantic memory (Ait Ameur and al., 2008;

Yurchyshyna et al., 2008) highlights the need for

query capitalization. Beyond query capitalisation, the

capitalisation of whole search processes becomes a

real challenge in many domains. By search process,

we mean a sequence of queries enabling to ﬁnd com-

prehensive and accurate information by composing

results from different information sources. These pro-

cesses are often difﬁcult to acquire by novice users

and they become more and more critical because

of the specialization and proliferation of knowledge

sources (Bhavnani et al., 2003).

(Bhavnani et al., 2003) presents an approach to

formalize critical search procedures in the medical

domain into strategy hubs. A search procedure is

represented by an ordered set of sub-goals and strat-

egy hubs provide search procedures and associated

high-quality links to information sources that enable

users to ﬁnd comprehensive and accurate informa-

tion. (Buffereau et al., 2003) propose to support nav-

igation among resources by e-road maps which are

composed of steps characterized by an intention or ti-

tle, an optional subject and optional illustrations (web

ressources). In this paper we propose a model to cap-

italize, reuse and share search queries and to orga-

nize them into formalized search processes. The orig-

inality of our proposal with regards to both strategy

hubs and e-road maps relies in that we statically as-

sociate resource patterns to steps of search processes.

Resources are dynamically selected when rendering

queries associated to search process sub-goals.

Our paper is organized as follows. Section 2

presents the concept of semantic search process and

the model we propose. Section 3 presents how search

processes are operationalized by rules. We conclude

in Section 4.

2 SEMANTIC SEARCH PROCESS

We deﬁne the notion of search process as a se-

quence of atomic searches to be processed by a do-

main expert to fulﬁll a domain-speciﬁc task or pro-

339

Corby O., Faron-Zucker C. and Mirbel I. (2009).

IMPLEMENTATION OF INTENTION-DRIVEN SEARCH PROCESSES BY SPARQL QUERIES.

In Proceedings of the 11th International Conference on Enterprise Information Systems - Artiﬁcial Intelligence and Decision Support Systems, pages

339-342

DOI: 10.5220/0001993103390342

 SciTePress

cess. A search process may be seen as a particu-

lar kind of business process limited to search activi-

ties. Different business process modeling formalisms

have been proposed in the literature. They can

be classiﬁed into three categories: activity-oriented,

product-oriented and decision-oriented ones (Nurcan

and Edme, 2005). Decision-oriented models are se-

mantically more powerful than the two others because

they explain not only how the process proceeds but

also why. Their enactment guide the decision making

process that shapes the process, and helps reasoning

about the rationale (Nurcan and Edme, 2005).

To support knowledge transfer about search pro-

cess from experts to novices, we are concerned with

the modeling of why the search process is decom-

posed the way it is, as well as with the speciﬁcation

of how it is decomposed. Moreover, to handle differ-

ent users’proﬁles and levels of knowledge, we want to

provide means to specify search processes at different

levels of detail. For all these reasons, the approach

we propose to model search processes is based on the

adaptation of an intentional process modeling formal-

ism (Rolland, 2007).

2.1 Maps

According to (Rolland, 2007), a map is a process

model in which a non-deterministic ordering of in-

tentions and strategies has been included. In our case,

we focus on search intentions and search strategies.

A map is a labeled directed graph with intentions as

nodes and strategies as edges between intentions. A

search intention is a goal that can be achieved by fol-

lowing a search strategy. An intention expresses what

is wanted, a state or a result that is expected to be

reached disregarding considerations about who, when

and where. There are two distinct intentions that rep-

resent the intentions to start and to stop the process.

A map consists of a number of sections each of which

is a triple (source intention, target intention, strategy).

A strategy characterizes the ﬂow from the source in-

tention to the target intention and the way the target

intention can be achieved. A map contains a ﬁnite

number of paths from its start intention to its stop in-

tention, each of them prescribing a way to achieve the

goal of the search process under consideration.

Let’s take the example of a teacher belonging to

a community of teachers and looking for resources to

build a course about relational database. To fulﬁll this

goal, s/he may search for resources about database

history to build the introduction of its course, then re-

sources about the relational model, normal forms and

SQL, and ﬁnally resources on how to interact with

a database from a programming language. Search

for resources about database history and search re-

sources about relational model are examples of in-

tention. Depending on the context of the course

(the number of teaching hours, the audience, ...), the

teacher may be or not interested by some resources,

for instance resources about database history. Conse-

quently, this intention looks optional and two paths

are suggested to fulﬁll the main goal of building a

course about relational database with or without the

Searchfor resources about database history intention.

The formalization of this scenario with the map model

is presented in ﬁgure 1(a).

Figure 1: Example of a Map.

As shown in ﬁgure 1(a), there might be several

ﬂows from a source intention to a target intention,

each correspondingto a speciﬁc strategy. It is the case

of intention

labeled by search for resources on how

to interact with a database from a programming lan-

guage which is reachable via a strategy based on the

Java programming language or via a strategy based on

the PHP programming language. There might also be

several strategies to reach a target intention from dif-

ferent source intentions. It is the case of intention

labeled by search for resources about the relational

model which is reachable from intention

labeled

by search for resources about database history or di-

rectly from the start intention.

Still according to (Rolland, 2007), the execution

of each section of a map is supported by an intention

achievement guideline (IAG) which provides an op-

erational or an intentional means to fulﬁll the target

intention. In our work, we operationalize an IAG by

the execution of a query on the community semantic

memory or we deﬁne it by a reﬁned map, as shown in

ﬁgure 1(b) and (c).

To further formalize intentions and strategies, we

rely on (Prat, 1999) proposal, which has already

proven to be useful to formalize goals (Ralyte, 2001;

ICEIS 2009 - International Conference on Enterprise Information Systems

340

Rolland, 2007). According to (Prat, 1999), an inten-

tion statement is characterized by a verb and some

parameters which play speciﬁc roles with respect to

the verb. Among the parameters, there is the object

on which the action described by the verb is pro-

cessed. In (Ralyte, 2001; Rolland, 2007) different

other relevant parameters have been identiﬁed : ben-

eﬁciary, reference, quality, quantity, direction, time,

ways (manner or means) and location. Let us consider

again the map depicted in ﬁgure 1(b). Intention

i3a

labeled by search for resources about normal forms

deﬁnition is described by its verb search and its ob-

ject resources about normal forms deﬁnition. Inten-

tion

labeled by search for resources on how to in-

teract with a database from a programming language

achieved through the PHP strategy is described by its

verb search, its object API and its manner PHP.

2.2 Search Process Modelling

Starting from the map model, we propose a search

process modelling based on Semantic Web standards.

We gathered the concepts and relationships of the map

model and we built an RDFS ontology dedicated to

the representation of search processes (including for

instance classes Verb, Object and Parameter to repre-

sent intention statement).

In order to specify intention statements, we are

currently exploiting a single verb, namely class

Search which instantiates class Verb, and we consider

many domain-speciﬁc concepts as instantiations of

class Object. The class Parameter is instantiated into

several classes modeling the context of the search pro-

cess. We distinguish between domain dependent and

domain independent contextual information. To fur-

ther populate the search process ontology with classes

of parameters, a domain independent context ontol-

ogy is under development and mappings with con-

cepts from a domain-speciﬁc ontology as well.

Based on the search process ontology, search pro-

cesses (or fragments of search process) are then rep-

resented by RDF annotations.

By relying on RDF(S) which is now a widespread

Web standard, we ensure the capitalization, reuse

and share of these representations of search processes

among community members. Beyond an alternative

way to organize and to dynamically access resources

in a community memory, we provide means to cap-

italize search processes themselves. We take advan-

tage of the inference capabilities provided by the RDF

framework to reason on search process representa-

tions, especially to organize them and retrieve them

for reuse.

3 INTENTION ACHIEVEMENT

GUIDELINE MODELLING

Reusing search processes (or search process frag-

ments) is intended to enable a dynamic connection

of different search processes and therefore the build-

ing of a whole search process by combining those

(fragments of) search processes which both satisfy the

global intention and retrieve available resources. This

goes through modeling IAGs which connect a section

of a map representing a search process either with

another map representing another search process (in-

tentional means) which can be viewed as a fragment

of the global process fulﬁlling the target intention of

the connected section, or directly with a query (op-

erational means) retrieving relevant resources in the

community semantic memory.

We propose to represent an IAG by a rule which

conclusion represents a section of a map and which

premise represents either an operational means (a

query) or an intentional means (a map) fulﬁlling the

target intention of the section in conclusion. We call

a rule concrete or abstract depending on wether its

premise represents operational or intentional means.

The SPARQL language provides a uniﬁed framework

to represent both concrete and abstract rules through

the CONSTRUCT query form. A CONSTRUCT query

form returns an RDF graph speciﬁed by a graph tem-

plate in the query form and constructed by taking

each query solution, substituting for the variables in

the graph template and combining the resulting RDF

triples. In our case we formalize a rule representing

an IAG by a SPARQL query which CONSTRUCT clause

is the conclusion of the rule, i.e. the graph template to

construct the RDF representation of a section of a map

(the conclusion of the rule) and which WHERE clause

is the premise of the rule, i.e. a graph pattern repre-

senting a map (abstract rule) or criteria for retrieving

relevant ressources (concrete rule).

In our running example, let us consider again the

IAG associated to the section of the map presented in

ﬁgure 1 aiming at searching for resources about nor-

mal forms. This IAG describes an intentional means

to fulﬁll target intention

of the section. We repre-

sent it by an abstract rule formalized by the following

SPARQL query, where preﬁx

map

refers to the search

process ontology and preﬁx

refers to a domain on-

tology on relational database:

CONSTRUCT { (1)

_:s map:hasTarget _:i (2)

_:i map:hasObject d:NormalForm (3)

_:s map:operationalizedBy ?g (4)

} (5)

WHERE { (6)

IMPLEMENTATION OF INTENTION-DRIVEN SEARCH PROCESSES BY SPARQL QUERIES

341

GRAPH ?g { (7)

?s1 map:hasSource ?i0 (8)

?i0 rdf:type map:Start (9)

?s1 map:hasTarget ?i1 (10)

?i1 map:hasObject d:NormalFormDef (11)

?s2 map:hasSource ?i1 (12)

?s2 map:hasTarget ?i2 (13)

?i2 map:hasObject d:NormalFormTransform

?s3 map:hasSource ?i2 (15)

?s3 map:hasTarget ?i4 (16)

?i4 rdf:type map:Stop (17)

}

The CONSTRUCT clause of this query is a graph

template for building an RDF graph representing any

section aiming at searching for resources about nor-

mal forms. It includes both statements (lines 2-3) de-

scribing the object of the target intention of the sec-

tion with the domain concept NormalForm instantiat-

ing concept Object of the map ontology and a state-

ment (line 4) about the RDF graph operationalizingthe

section and which content is described in the WHERE

clause of the query. This links together the two levels

of intention reﬁnement.

The WHERE clause of the query describes how to

operationalize any section (in particular the one of our

example) whose RDF representation matches with the

graph template in the CONSTRUCT clause. It is a

graph template that matches with the RDF represen-

tation of the map shown in ﬁgure 1. It includes state-

ments about three sections: the ﬁrst ones (lines 8-11)

describe a ﬁrst section

?s1

which source intention is

a start and which target intention has for object the

deﬁnition of a normal form; the following ones (lines

12-14) describe a second section

?s2

which source

intention is the target intention of the ﬁrst section

?s1

and which target intention has for object the transfor-

mation of normal forms; the last ones describe a third

section which source intention is the target intention

of the second section

?s2

and which target intention

is a stop.

Instantiating search processes, i.e. combining

sub-processes into a global process is achieved by ap-

plying rules implementing IAGs in backward chain-

ing. The problem of operationalizing the initial strat-

egy provided to the backward chaining engine then

boils down to operationalizing the sections described

in the WHERE clause of the query. We rely on the

CORESE

(Corby et al., 2006) semantic engine for

both backward chaining on the knowledge base of

SPARQL queries and matching whith the knowledge

base of RDF annotations of domain resources.

http://www-sop.inria.fr/edelweiss/software/corese/

4 CONCLUSIONS

In this paper, we proposed an approach relying on

semantic Web technologies and models to capital-

ize, reuse and share search queries and search pro-

cesses. By modeling search processes, our aim was

to capture knowledge and best practices into series

of structured search activities. Therefore, starting

from an intention driven process modeling formal-

ism, we proposed an ontology to annotate search pro-

cesses and we operationalized guidelines associated

to search processes fragments with rules implemented

as SPARQL queries. As a result, instantiation of search

processes is supported by backward chaining among

the rule base and matching with the RDF dataset an-

notating the community resources.

REFERENCES

Ait Ameur, Y. and al. (2008). Semantic hubs for geograph-

ical projects. In Semantic Metadata Management and

Applications (SeMMA), workshop at ESWC.

Bhavnani, S., Bichakjian, C., Johnson, T., Little, R., Peck,

F., Schwartz, J., and Strecher, V. (2003). Strategy

hubs: Next-generation domain protals with search

procedures. In ACM Conference on Human Factors

in Computing Systems.

Buffereau, B., Duchet, P., and Picouet, P. (2003). Gener-

ating guided tours to facilitate learning from a set of

indexed resources. In IEEE International Conference

on Advanced Learning Technologies (ICALT),Athens,

Greece.

Corby, O., Dieng-Kuntz, R., Faron-Zucker, C., and Gandon,

F. (2006). Searching the semantic web: Approximate

query processing based on ontologies. IEEE Intelli-

gent Systems Journal, 21(1).

Nurcan, S. and Edme, M. (2005). Intention-driven modeling

for ﬂexible workﬂow applications. Software Process:

Improvement and Practice, 10(4):363–377.

Prat, N. (1999). R´eutilisation de la trace par apprentissage

dans un environnement pour l’ing´enierie des proces-

sus. PhD thesis, Universit´e Paris I - Sorbonne.

Ralyte, J. (2001). Ing´enierie des m´ethodes `a base de com-

posants. PhD thesis, Universit´e Paris I - Sorbonne.

Rolland, C. (2007). Conceptual Modelling in Information

Systems Engineering, chapter Capturing System In-

tentionality with Maps. Springer-Verlag.

Yurchyshyna, A., Faron-Zucker, C., Mirbel, I., Sall, B.,

Le Thanh, N., and Zarli, A. (2008). Une approche

ontologique pour formaliser la connaissance experte

dans le mod`ele du contrˆole de conformit´e en construc-

tion. In 19i`eme journ´ees francophones d’ing´enierie

des connaissances, Nancy, France.

ICEIS 2009 - International Conference on Enterprise Information Systems

342