Prefer-W2S: Preferences-Aware Query in Web Services Selection

Transcription

Prefer-W2S: Preferences-Aware Query in Web Services Selection
Prefer-W2S: Preferences-Aware Query in Web
Services Selection
Zahira Chouiref , Abdelkader Belkhir and Allel Hadjali
Bouira University, Departement of Computer Science, Algeria/
zahirachouiref@univ-bouira.dz
USTHB University, 16000 Algiers, Algeria/kaderbelkhir@hotmail.com
Universit´e de Poitiers - ISAE-ENSMA, LIAS lab, France/allel.hadjali@ensma.fr
Abstract. In this paper, we discuss how Web services selection can be
highly improved by leveraging both explicit and implicit user preferences.
The former are modeled in a convenient way by means of fuzzy sets.
The latter which depend on information related to the user profile and
context, are discovered thanks to an appropriate fuzzy inference schema.
Resulting services are more customized and fit better the user needs. The
first experiments done show the usefulness of the proposal.
Keywords: Web Services Selection, Preferences, Fuzzy Logic Theory.
1
Introduction and related work
Nowadays, vast repositories that contain huge amounts of Web services (WS )
on all matters of interest, are available. Also, large corporations are now founding their business on an abundant use of WS, the number of publicly available
services is then envisioned to increase in the future [5]. Therefore, there will be
a large number of candidate services for a desired task. Non-functional properties of WS are a key concept that can be used to rank these candidate services
according to user preferences and to select the best ones. The research literature on preferences in the context of WS selection is extensive. It encompasses
preference-aware quality of service (QoS ), context-aware preferences, etc. In [8],
authors proposed a list of non-functional properties that are related to user preferences and context of the demanded WS. However, the model proposed by the
authors does not consider the implicit user preferences and could not handle both
exact and imprecise/fuzzy requirements. Chao et al. [3] proposed a framework
which leverages fuzzy logic to abstract and classify the underlying data of WS in
terms of fuzzy labels and rules. The aim is to allow the use of vague terms at the
level of the search query. Adnan et al. [1] proposed a matchmaking algorithm
to discover WS that are satisfying client requirements. The aim of the paper is
tying QoS metrics of WS with fuzzy words that are used in users request. The
shortcoming of the paper is to satisfy user’s preferences regarding only QoS. In
[9], a multi-phase matching is proposed, which exploits WS capability (Input
and Output) and QoS’s fuzzy information of WS (e.g. cost). The authors do not
2
Zahira Chouiref , Abdelkader Belkhir and Allel Hadjali
take into account the vague/gradual nature of information related to user profile
and context. See also the work by [7] based on rough sets theory. Although many
methods for WS selection have been developed, there is a lack of a generic and
flexible preference model that allows to define (in a faithful way) user preferences related to the relevant services to be selected. In sharp contrast to the
existing WS selection approaches, that focus only on user-specified preferences,
in this paper, we propose a flexible and effective WS selection framework, which
provides users with an adequate way to express their preferences using linguistic
terms, and enhance the service selection by the integration of the context and
the profile of the user. According to his/her contextual profile (CP ) [4], a user
may have several preferences. For instance, consider a user that wants to book a
restaurant in weekend . (S)He sets his/her preferences and submits the following query Q: ”return the restaurants preferably with {affordable price} and
{french cuisine}, knowing that (s)he smoke, (s)he is conveyed and (s)he is
accompanied by his/her family . Users are then more satisfied by restaurants
that take their contexts (e.g. accompanying people, means of transport, etc.) and
their profiles (e.g. profession, age, etc.) into account. Given a request, we introduce an objective score to measure to which extent a retrieved service is relevant.
This score takes into account no only the user-specified preferences, but also additional preferences extracted from both his/her context and profile using fuzzy
rules, so as to improve the effectiveness of the selection. In this context, the service selection and ranking process are accomplished by the integration of three
dimensions: user profile, user context and explicit/implicit user preferences into
the end user querying process. Explicit preferences are provided in the original
user query. Implicit preferences are inferred from user CP and used to expand
the original query in order to retrieve more relevant services to the same purpose.
So, our framework makes use of an appropriate fuzzy inference mechanism to
precisely capture the implicit preferences for ranking the best services. In Section 2, we set up a CP and preferences model, then we present an advanced WS
selection framework in Section 3. In Section 4, first results of an experimental
study are described. Finally, Section 5 concludes the paper and outlines some
future work.
2
2.1
Contextual Profile and Preference Model
Modeling Contextual Profile
We define a CP = (hcontexti, hprof ilei) of user and service as a couple of context concepts and profile concepts. This set of concepts, called contextual profile
parameters (Pi ), is stored in a hierarchical data structure called contextual profile
ontology, as shown in Fig. 1. For a given query Q (service S resp.), we define its
CP environment CP EQ (CP ES resp.) as a set {(P1 , . . . Pn )} of n multidimensional parameters. In general, CP environment can be distinguished in different
ways. Hence, it is important to have a clear view of how such information would
be defined in a given domain. For instance, the user CP environment of our
example is {Accompanying people, Time period, Means of transport, Profession,
Lecture Notes in Computer Science: Authors’ Instructions
3
Fig. 1: User/Service Contextual Profile Ontology
Demographic information}, and the service CP environment is {Time period,
Facilities}. Each parameter (Pi ) can be represented by a set of atomic concepts
or composed concepts {(C1 , . . . Cm )}, for instance, atomic concept: Accompanying people = {Alone, Family, Friends}, composed concept: Time period =
{Season, Date, Time}, and Time = {Morning, Night, Evening, Weekday, Weekend}. Thus, each concept Ci takes its values from a hierarchical domain, called
extended domain, edom(Ci ). An instantiation of the CP, called CP state, writes:
w = (C1 is v1 ∧ ... ∧ Ck is vk ), k ≤ m , where each Ci ∈ CP EQ , 1 ≤ i ≤ k
and vi ⊆ edom (Ci ) (the symbol ∧ denotes a conjunction). For instance, w may
be (Accompanying people is family ∧ Time is weekend ∧ Means of transport is
cars ∧ Profession is salaried ∧ Age is young couple) for our previous query.
2.2
CP-dependent Preferences
A contextual profile preference (CPP) is a fuzzy rule of the form: if C1 is v1
∧ ... ∧ Cm is vm then A1 is F1 ∧ ... ∧ Al is Fl , where vi , 1 ≤ i ≤ m ≤
n, stands for a crisp or fuzzy value of the context or the profile parameter CPi
and Fj , 1 ≤ j ≤ l represents a fuzzy preference related to attribute Aj . The
meaning of CPP is that in the CP state specified by the left part of the rule,
the preference ”Aj is Fj ” is inferred. For instance, from the observation that a
user has a car, a preference for a restaurant with parking is of interest. This may
be expressed as (CP P1 ): if means of transport is car then PreferenceParking
is yes. Note that a fuzzy value of numerical attribute (e.g. ”affordable price” )
is described thanks to fuzzy sets, allowing to obtain a satisfaction level in the
interval [0, 1]. As for categorial attributes, ”CuisineStyle” is for instance modeled
by: F cuisine={1/French, 0.9/Belgian, 0.7/Italian, 0.7/Dutch, 0.4/British} for
French cuisine.
3
An Advanced Web Services Selection Framework
We describe here a selection framework that leverages domain knowledge and
fuzzy inference technology to enhance the quality of an existing WS selection
engine. Fig. 2 gives an overview of this selection framework. We focus mostly
4
Zahira Chouiref , Abdelkader Belkhir and Allel Hadjali
Fig. 2: An Advanced Web Service Selection Framework
on explicit user preferences. Such preferences can refer only to built-in query
predicates by requester. In that way we match preferences that are based only on
the values occurring in query attributes with WS parameters (step 1 and 2). To
compute the satisfaction degrees of retrieved services w.r.t to explicit preferences,
for example for Price and CuisineStyle, expressed in a fuzzy way, we use their
memberships functions. For instance, one can obtain f1 = µaf f ordable (45) = 0.5
and f2 = µF Cuisine (Dutch) = 0.7. It is worth noticing the all the degrees
f1 , f2 belong to the same scale [0,1]. This property of commensurability allows
aggregating them, in a convenient way, to obtain an overall score Score1 of a
service using a T-norm operator (e.g., the min operator ) as follows: Score1 =
>(f1 , f2 ) = min(f1 , f2 ). Now, if the Score1 of a service is higher than P
θ (a
user-defined threshold), then this service is added to the list of services ( Q )
that satisfy explicit user preferences. As for implicit preferences, steps (3, 3’,
4, 5, 5’, 6) illustrate how they can be used in our selection process and are
summarized in the following: (i) Inferring a set of relevant preferences from
the fuzzy rules base B CP P 1 , regarding the user CP state w, then augment the
query by the inferred preferences. To achieve this, we make use of a knowledge
based model described bellow. (ii) Computing of the satisfaction of the result
provided by steps 1 and 2 w.r.t. the inferred preferences. One assume available
B CP P = {CP P1 , ..., CP Pm } a fuzzy rules base modeling a set of CP preferences.
Now, given a user’s CP state, one can derive relevant preferences to the user
by using a fuzzy inference schema, called the Generalized Modus Ponens (GMP)
[6]. In a simple case, from the rule: if C is V then A is F and the fact: C is V’,
where V, F and V’ are gradual predicates modeled thanks to fuzzy sets, GMP
allows inferring the preference A is F’ and the fuzzy semantics of F’ (see [6]
for more details). For instance, the preference ambience is animated’ is inferred
by applying the rule (If age is young then ambience is animated ) and the fact
(age is about 27 ). Note that the fuzzy semantics of the predicate animated’ is
computed using the semantics of animated, young and about 27, by means of
1
is a set of contextual profile preferences that can be built from users’ experiences in
the domain considered.
Lecture Notes in Computer Science: Authors’ Instructions
5
the combination/projection principle [2]. Once the implicit user preferences are
inferred, the system augments the query by these preferences in order to refine
the selection process. Then,
the system evaluates the satisfaction degree, Score2 ,
P
of each service WS of Q w.r.t. the (fuzzy) implicit preferences. Finally, we use
an aggregation P
function of Score1 and Score2 to compute the overall score S
of each WS of Q . To give priority to the initial preferences w.r.t. to inferred
preferences (IPS) we make use of the following formula (where α ∈]0, 1[ is the
priority of IPS): S = min(Score1 , max(Score2 , 1 − min(Score1 , α))). One can
observe that the score S behaves as an ordinary conjunction for high values of
Score1 and Score2 , while S = Score1 when Score1 is below some threshold
1 − α. Thus, the one can select the top-k answers or the answers whose score S
is greater than a given θ.
4
Experimental Evaluation: First Results
The main purpose of this evaluation is to compare the effectiveness of our proposed selection framework (referred to as IP for inference process) with the
traditional frameworks that do not use the inference process (referred to as TR
for traditional). We created a set of 100 synthetic restaurant service descriptions,
and we involved different users to conduct our experiments. However, due to lack
of space we only report results regarding 4 users. Fig. 3 shows the precision of IP
and TR at various ranks for 4 different users. Observe that IP has consistently
better precision than TR since IP includes into the ranking process inferred preferences that are very interesting for users. See also that, for user1 and user2 IP
has an almost perfect precision, while the precision of TR is mediocre. Moreover,
for user3 and user4 IP and TR have similar precision at rank 15 and rank 20.
The reason is that the increase of the rank may increase the probability that
similar services belong to the top-k list of both approaches.
5
Discussion and Conclusion
Handling user preferences is becoming an increasingly important issue in presentday services search. In this paper, we have discussed a step to cope with this issue
in an efficient way. Taking into account both explicit and implicit preferences
in the selection process, is the key idea of the approach proposed. Thus, delivered services, which fit better users’s requirements, are rank-ordered in a more
practical and useful way. The first experiments demonstrate really the interest
of the approach. As for future works, we plan to conduct thorough experiments
and address the case where the user context is pervaded with uncertainty.
References
1. Al Rabea, A.I., Al Fraihat, M.M.: A new matchmaking algorithm based on multilevel matching mechanism combined with fuzzy set. Journal of Software Engineering
and Applications 5(3) (2012)
6
Zahira Chouiref , Abdelkader Belkhir and Allel Hadjali
1
1
IP
TR
IP
TR
0.8
Precision
Precision
0.8
0.6
0.4
0.2
0.6
0.4
0.2
0
0
5
10
15
20
5
10
Rank
(a) user1
20
1
(b) user2
1
IP
TR
IP
TR
0.8
Precision
0.8
Precision
15
Rank
0.6
0.4
0.2
0.6
0.4
0.2
0
0
5
10
15
Rank
(c) user3
20
5
10
15
20
Rank
(d) user4
Fig. 3: Precision at Rank (θ = 0.5, α = 0.5)
2. Bouchon-Meunier, B., Dubois, D., Godo, L., Prade, H.: Fuzzy sets and possibility theory in approximate and plausible reasoning. In: Fuzzy sets in approximate
reasoning and information systems, pp. 15–190
3. Chao, K.M., Younas, M., Lo, C.C., Tan, T.H.: Fuzzy matchmaking for web services.
In: Advanced Information Networking and Applications, 2005. AINA 2005. 19th
International Conference on. vol. 2, pp. 721–726. IEEE (2005)
4. Chouiref, Z., Belkhir, A., Hadjali, A.: Enhancing semantic web services discovery
using similarity of contextual profile. In: ICIW 2013, The Eighth International Conference on Internet and Web Applications and Services. pp. 95–99 (2013)
5. Davies, J., Potter, M., Richardson, M., Stinˇci´c, S., Domingue, J., Pedrinaci, C.,
Fensel, D., Gonz´
alez-Cabero, R.: Towards the open service web. BT Technology
Journal 26(2) (2009)
6. Hadjali, A., Mokhtari, A., Pivert, O.: A fuzzy-rule-based approach to contextual
preference queries. In: Computational Intelligence for Knowledge-Based Systems
Design, pp. 532–541. Springer (2010)
7. Pokhrel, J., Lalanne, F., Cavalli, A., Mallouli, W.: Qoe estimation for web service
selection using a fuzzy-rough hybrid expert system. In: 2014 IEEE 28th Int. Conf.
on Advanced Inf. Networking and App. (AINA). pp. 629–634. IEEE (2014)
8. Reiff-Marganiec, S., Yu, H.Q.: An integrated approach for service selection using non-functional properties and composition context. Handbook of Research on
Service-Oriented Systems and Non-Functional Properties: Future Directions pp.
165–191 (2011)
9. Su, Z., Chen, H., Zhu, L., Zeng, Y.: Framework of semantic web service discovery
based on fuzzy logic and multi-phase matching. Journal of Information and Computational Science 9, 203–214 (2012)