WEB SERVICES & RECOMMENDER SYSTEMS

A Research Roadmap

Leandro Krug Wives, José Palazzo Moreira de Oliveira

Universidade Federal do Rio Grande do Sul, Brazil

Zakaria Maamar

Zayed University, U.A.E.

Samir Tata, Mohamed Sellami

TELECOM & Management Sud Paris, France

Keywords: Web services, Recommender systems, Research roadmap.

Abstract: In this position paper, we provide a brief overview of Recommender Systems (RS) and Web Services (WS).

After, we propose a research roadmap for the challenges and opportunities that could arise following the

combined use of WS and RS. While these challenges are expected to hinder this use, we discuss the neces-

sary actions that need to be taken to overcome these challenges and hence, make this use a win-win situation

for both WS and RS. We illustrate how the combination of RS to WS takes place in terms of what RS can

do for WS and what WS can do for RS. Finally, we conclude by pointing out the actions to take so that this

combination turns out successful.

1 INTRODUCTION

The reason of proposing a research roadmap for the

Web Services & Recommender Systems association

is the following. On the one hand, Web services'

adoption is somehow slowed down by various, re-

current issues largely reported in the literature such

as the complexity of Web services discovery and

Web services reliability (Sapkota 2005; Shen and Su

2007; Margaria 2008). To address these issues, in-

novative solutions are required and could be built

upon different other techniques for instance recom-

mendation (i.e. recommender systems). On the other

hand, recommender systems are slowly moving from

simple applications (e.g., textbook recommendation)

while modern business application's complexity

continues to increase. This has, to a certain extent,

excluded recommender systems from the list of

techniques of choice for the development of these

applications, unless the way recommender systems

are developed sees some changes. These changes

will permit to enhance recommender systems with

new capabilities drawn from existing IT technolo-

gies for instance Web services.

Web services are paving the way for a new gen-

eration of large-scale, loosely-coupled business

applications. This is witnessed from the large num-

ber of standards and projects related to Web ser-

vices, e.g., (Bentahar, Maamar et al. 2007; Yu,

Bouguettaya et al. 2008), which address a variety of

issues such as semantics, high availability, and dis-

covery. These issues, in fact, hinder the smooth

automatic composition and deployment of Web

services. Composition, which is one of Web ser-

vices' selling points, handles the situation of a user's

request that cannot be satisfied by any single, avail-

able Web service, whereas a composite Web service

obtained by combining available Web services may

be used.

Recommender systems are a special kind of in-

formation filtering approaches designed to help

users cope with information overload (Adomavicius

and Tuzhilin 2005; Werthner, Hansen et al. 2007).

These systems can process large volume of informa-

tion prior to suggesting items to a user according to

119

Krug Wives L., Palazzo Moreira de Oliveira J., Maamar Z., Tata S. and Sellami M.

WEB SERVICES RECOMMENDER SYSTEMS - A Research Roadmap.

DOI: 10.5220/0002840601190124

In Proceedings of the 6th International Conference on Web Information Systems and Technology (WEBIST 2010), page

ISBN: 978-989-674-025-2

first, the users’ interests or behaviour and second, to

other users’ past experiences and recommendations

(e.g., comments, criticisms, and opinions). This is

very appealing when users are overloaded with in-

formation and searching for relevant ones turns out a

“nightmare”. Thanks to recommender systems, the

search space could be reduced (in the user's perspec-

tive), reducing the number of elements the user has

to analyze, providing a positive impact on the com-

plexity and time needed to find relevant information,

e.g., (Pu, Chen et al. 2008).

In the remaining of this position paper, we pro-

vide a brief overview of Recommender Systems

(RS) and Web Services (WS). Afterwards, we illus-

trate how the combination of RS to WS takes place

in terms of what RS can do for WS and what WS

can do for RS. Finally, we conclude by pointing out

the actions to take so that this combination turns out

to be successful.

2 RECOMMENDER SYSTEMS

AND WEB SERVICES

According to Werthner et al. (2007), recommender

systems are applications that provide qualified ad-

vices about products or services that a user might be

interested in, though the final word of selecting that

product or that service over others goes always back

to the user. In addition, RS can assist people make

the right decisions when they are swashed with an

overwhelming set of alternatives.

There exist several commercial and academic

applications that illustrate the use of RS. Examples

include Amazon

, Last.fm

, Phoaks

iii

system, and all

the Grouplens’ projects

(MovieLens, WikiLens,

TechLens, etc.). RS can be classified into three cate-

gories: content-based, collaborative filtering, and

hybrid. This classification takes into account the

recommendation model upon which a RS is built. A

good survey on RS is proposed in (Adomavicius and

Tuzhilin 2005) and (Perugini, Gonçalves et al.

2004).

 Content-based Recommendation: here, a rec-

ommendation is performed by analyzing the

similarity between (i) items (artefacts) and a

user’s interests (represented by her profile in

terms of interests, level of expertise, and

sometimes needs) or (ii) items themselves.

The items that are more similar to the user’s

profile are suggested first. The similarity iden-

tification algorithm depends on the character-

istics like price, size, and purpose of the item

that is being analyzed. In addition, the similar-

ity functions that are used in these algorithms

depend on the nature of the item.

 Collaborative Filtering Recommendation:

here, users explicitly evaluate items they know

or use by giving specific scores to these items

(Herlocker, Konstan et al. 2004). Because it is

known that users are always reluctant to en-

gage in any scoring exercise (Claypool, Le et

al. 2001), other alternatives exist to address

this reluctance and rely, for example, on track-

ing how a user interacted with the system by

registering the items she bought, which is

similar to what Amazon does. An item could

be any object, document, product, or service.

The highly evaluated items are recommended

to users using neighbourhood-based algo-

rithms. The main idea is that users with simi-

lar profiles tend to have same interests. The

similarity among items is also important to

identify since similar items may be relevant to

users with similar interests. The major limita-

tion of collaborative filtering recommendation

is that users do not receive recommendations

until some evaluation is performed by another

user.

 Hybrid Recommender Systems: Melville et al.

(2002) state that content-based and collabora-

tive filtering recommendations “fail to provide

good recommendations in many situations”.

Thus, their combination seems to be a good

way to address the limitations of each recom-

mendation type. This is where hybrid RS

came into play. There exist many approaches

to combine RS techniques. Melville et al., for

instance, use content-based techniques to

identify similarities among rated- and unrated-

items, which permits to minimize the lack of

evaluation that was mentioned before. Other

authors apply different Recommendation

Techniques (RT) and combine or choose their

better results (Ahmad Wasfi 1999; Torres,

McNee et al. 2004; Thio and Karunasekera

2005). While there is not a complete study yet

that states which combination approach to go

with, existing studies show that hybrid RS

provide better results (Adomavicius and

Tuzhilin 2005).

Web service as defined by the W3C is “a soft-

ware application identified by a URI, whose inter-

faces and binding are capable of being defined, de-

scribed, and discovered by XML artefacts, and sup-

ports direct interactions with other software applica-

tions using XML-based messages via Internet-based

WEBIST 2010 - 6th International Conference on Web Information Systems and Technologies

120

applications” (W3C 2004). A WS implements a

functionality (e.g., BookOrdering) that users and

other peers (i.e., Web services) invoke by submitting

appropriate messages. WS are modular and loosely-

coupled, providing a simple model of programming

and deployment of applications through the Web.

The life cycle of a WS consists of three steps usually

known as definition and announcement, discovery,

and invocation. A fourth step, which is about com-

position, is usually to this life cycle.

 Definition & Announcement Step: independ-

ent providers use the Web Service Description

Language (WSDL) to describe the functional-

ities, operations, and attributes of the WS they

develop. Afterwards, the providers announce

the Web services to the external community

by posting their descriptions on various public

and private registries (e.g., Universal Descrip-

tion Discovery and Integration (UDDI)

(OASIS 2004), Electronic Business using eX-

tensible Markup Language (ebXML) (OASIS

1999)).

 Discovery Step: finding WS means screening

and querying registries in order to match WS’

functionalities to service requesters. A query

includes search criteria in terms of functional

properties (type of service, inputs, outputs,

etc.) and sometimes non-functional properties

(response time, execution rate, etc.).

 Invocation Step: it is the process of calling the

operation of a WS. As software artefacts can

be developed using different languages, these

artefacts must follow a common contract to in-

teract with other elements in a platform-

neutral environment. In the case of WS, the

Simple Object Access Protocol (SOAP) is

used so that artefacts can be bound to each

other. SOAP is a XML-based remote invoca-

tion protocol designed for flexibly composing

Internet applications.

 Composition Step: when a user’s request can-

not be satisfied by any single, available WS,

the combination of WS constitutes an alterna-

tive (Berardi, Calvanese et al. 2003). Exam-

ples of specification languages to compose

Web services include the Business Process

Execution Language (BPEL) and the Web

Services Choreography Description Language

(WSCDL).

3 WEB SERVICES &

RECOMMENDER SYSTEMS

COMBINATION

In this section, we propose a research roadmap for

the challenges and opportunities that could arise

following the combined use of WS and RS. While

these challenges are expected to hinder this use, we

discuss the necessary actions that need to be taken to

overcome these challenges and hence, make this use

a win-win situation for both WS and RS.

3.1 What can RS do for WS?

Recommendation seems to offer solutions to the

problems of handling large volumes of data (Pu,

Chen et al. 2008). Users’ profiles, past experiences,

and rankings are some of the factors that RS take

into consideration in their functioning. With the

proliferation of WS-based applications, recommen-

dations would help address some recurrent issues of

how to discover WS with reduced efforts, how to

suggest/avoid “good/poor” WS based on previous

successful/unsuccessful executions, how to compare

WS, just to cite some. In this section, we discuss

how recommendations could be smoothly woven

into the life cycle of a WS.

With the proliferation of trusted and un-trusted

WS, their discovery (including registries upon which

these WS are posted) constitutes and continues to be

complex and tedious, although several improve-

ments are made to the operational mechanisms of

registries to address, for example, semantic and

security concerns (Trabelsi, Pazzaglia et al. 2006).

The discovery process should be performed based on

the semantic match between a declarative descrip-

tion of the WS that is being sought, and a description

of the WS that is being offered (Paolucci, Kawamura

et al. 2002), but this activity is still “complex” and

poses problems to allow a broader adoption of WS

by businesses. Recommendation could simplify this

discovery process by easing and speeding up some

of the activities that are related to one of the follow-

ing three cases:

 Case 1: instead of screening registries, which

is the traditional way of discovering WS, a RS

could suggest a list of WS that satisfies a

user’s or client's needs based on previous ex-

periences collected over time. With respect to

this user or client, these experiences should

feature similar needs, profiles, interests, func-

tionalities, etc. The value-added of RS to this

case is to reduce the search space of WS to a

WEB SERVICES & RECOMMENDER SYSTEMS - A Research Roadmap

121

specific list of recommended WS, which is the

main shortcoming in WS discovery.

 Case 2: in a composition scenario, a RS could

suggest additional WS to be included in this

scenario based on the components that are al-

ready included and the user’s profile or based

on the composition's functionality. For in-

stance, a delegate attending an overseas con-

ference could be interested in some sightsee-

ing activities, though these activities are not

part of the composition scenario associated

with this delegate’s travel plan. The new com-

ponent WS will be submitted to the user for

approval before inclusion and execution.

 Case 3: to make WS more robust, a RS could

suggest peers that would substitute this WS in

case of failure. The peers are recommended

based on the functional and non-functional

characteristics they have in common with the

WS to make robust.

WS discovery continues to rely on how they are

described. By weaving some Recommender Tech-

niques (RT) into this discovery, WS’ descriptions

could turn out insufficient, inefficient, and even

inappropriate. The WSDL document of a WS uses a

number of well-defined, XML-based arguments

(e.g., message, port type, binding) that are primarily

meant for discovery and invocation purposes. Unfor-

tunately, these arguments cannot sustain the normal

functioning of any RS that requires to know, for

example, how a WS was rated by users, by whom a

WS was recommended in the past, how many times

a WS was discovered through recommendation, how

many times a user rejected a WS despite positive

recommendations and vice versa, etc. As a result,

dedicated arguments to describe WS are required

and could be related to user/client evaluation (score)

over a certain time frame, reputation as defined by

users and peers, cases that failed/succeed despite

positive/negative recommendations, etc. To monitor

and log WS' status and messages would also be

helpful to determine its quality and reputation.

Once WS’ new descriptions permit meeting the

requirements of how RS function, the next step

consists of looking into how RTs would affect the

announcement of these WS. Some of the aspects to

consider include:

 Where are the recommendation details an-

nounced? Should these details be posted on

existing registries like any regular WSDL de-

tail, or should these details be separated and

posted on dedicated registries that would be

made accessible to RS only?

 Should all recommendation details be pub-

lished or just some? In the latter case, what are

the details to select?

 When are the announced recommendation de-

tails refreshed? Will this be done regularly,

continuously, or both? And who is responsible

of performing this refreshment?

Another point of what RS can do for WS is about

the value-added of recommendation to composition.

The purpose here is to promote the reuse of compos-

ite WS. Rather than developing composite WS from

scratch each user’s request is submitted, the user is

given the possibility of triggering a composite WS

that was developed in the past in response to the

requests of other users and hence, was experi-

enced/tested/evaluated over time. This permits to

build the history of a composite WS in terms of

needs satisfied, problems faced, solutions adopted,

and performance assessed. Similarity of needs, pro-

files, and interests is a pre-requisite to the reuse of

composite WS. Because we have pointed out that the

descriptions of component WS need extra details

that would support the functioning of RS, we expect

that composite WS would be subject to the same,

namely extra details would be required. These de-

tails are for the composition level and could show,

for example, competitors of composite WS, per-

formance of composite WS, participating compo-

nents in compositions, etc.

Recommendation would not be complete without

taking full benefits of previously executed compos-

ite WS in terms of outcomes produced, obstacles

met, exceptions raised, and alternatives adopted.

Business Process (BP) mining is commonly used for

tracking execution (van der Aalst, Reijers et al.

2007), and is a good source of data for recommenda-

tion.

3.2 What can WS do for RS?

We now discuss how current RTs could be subject to

enhancement or adaption in response to WS’ charac-

teristics. These characteristics show, for example,

how WS are described, discovered, invoked, and

composed.

Content-based RTs rely on how items to analyze

and then to suggest are described in terms of con-

tent. When it comes to WS, their descriptors (i.e.,

WSDL documents) are the content to be analyzed.

These documents are XML-based and contain spe-

cific metadata about WS. This means that WSDL

documents are semi-structured and use a set of pre-

defined tags that contain textual, categorical, and

numerical values. The following list shows how the

WEBIST 2010 - 6th International Conference on Web Information Systems and Technologies

122

nature of these tags would affect the actions to take

during recommendation.

 Compared to textual documents, WSDL

documents are more structured and present a

strict syntax and semantics. Furthermore,

WSDL documents do not have to cope (to a

certain extent) with the vocabulary issue

(morphological variations, synonymy, etc.)

that is usually present on textual documents

and difficult identification of similar content.

Therefore, it is expected that minor changes in

the current content-based RTs would happen.

These changes would primarily permit to han-

dle the hierarchical structure and semantics of

WS.

 WS do not usually have textual attributes (in

the sense of free and unstructured texts). As

WS description languages are usually a varia-

tion of XML with predefined tags and limited

set of attributes, their contents are like semi-

structured data and thus more subject to cate-

gorical or numerical manipulation techniques.

Categorical and numerical attributes are thus,

extracted out of the WSDL tags of a WS.

These tags could be semantically annotated

(using ontology) so that a meaning is given to

each tag. To simplify the development of a

recommendation model rather than going

through the “hassle” of adapting existing RTs,

it would be interesting to assign priorities and

weights to the different WSDL tags. This

would, for instance, help the newly developed

recommendation model in ranking WS with

respect to different dimensions such as price,

availability, stability, etc.

Collaborative-based RTs are based on user col-

laboration, i.e., users must submit feedback (e.g.,

opinions, ratings) about the items they use or con-

sume, and the system recommends items based on

users with similar interests and opinions. When it

comes to WS, RTs can be used to minimize the

search space using previous experiences. The re-

finements or adaptations that must be performed in

RTs are related to collecting data about the WS's life

cycle and their interaction (sequence, order of

use/activation, status, quality), using specific moni-

tors (Cruz, Campos et al. 2004), since users do not

directly interact with WS (in fact, most of the time

they even do not know they are using WSs).

Hybrid-based RTs must be designed along the

following dual goals: deal with WS’ descriptors

(purpose of content-based techniques) and take into

account the way WS interact and are used (purpose

of collaborative-filtering techniques). The former

goal is related to collecting feedbacks on, for in-

stance, the evaluation (or any other kind of feedback

such as the number of activations or uses) of the

recommended or used WS, from any independent

peer (preferably trusted) or entity involved in over-

seeing the progress of the life cycle of a WS. The

latter goal is related to the lack of details that fol-

lows the similarity analysis of descriptions of WS.

For this purpose, hybrid techniques would use de-

scriptions of WS and feedbacks that are collected

out of the bodies responsible for the monitoring of

these WS.

4 SUMMARY

In this position paper, a research roadmap for com-

bining Web services and recommender systems was

discussed. This roadmap offered a glimpse of the

research opportunities and challenges that such a

combination would offer. What Web services can do

for recommender systems and the other way around

permitted to suggest an overview of first, the solu-

tions that Web services could offer to address the

limitations of recommender systems and second, the

mechanisms that recommender systems could offer

to sustain the adoption of Web services. Doubtless

combining Web services and recommender systems

should be a win-win situation for both, and real

business applications that would benefit out of this

combination will just prove this situation.

ACKNOWLEDGEMENTS

This work was partially supported by

CAPES/COFECUB project ADContext, and by

CNPq – Conselho Nacional de Desenvolvimento

Científico e Tecnológico, Brazil.

REFERENCES

Adomavicius, G. and A. Tuzhilin (2005). "Toward the

Next Generation of Recommender Systems: A Survey

of the State-of-the-Art and Possible Extensions." IEEE

Trans. on Knowl. and Data Eng. 17(6): 734-749.

Ahmad Wasfi, A. M. (1999). Collecting user access

patterns for building user profiles and collaborative

filtering. Proceedings of the 4th international

Conference on intelligent User interfaces Los Angeles,

California, United States, ACM, New York, NY: 57-

64.

Bentahar, J., Z. Maamar, et al. (2007). "An Argumentation

WEB SERVICES & RECOMMENDER SYSTEMS - A Research Roadmap

123

Framework for Communities of Web Services." IEEE

Intelligent Systems 22(6): 75-83.

Berardi, D., D. Calvanese, et al. (2003). A Foundational

Vision for e-Services. Proceedings of Ubiquitous

Mobile Information and Collaboration Systems

Workshop (UMICS’2003) held in conjunction with

The 15th International Conference On Advanced

Information Systems Engineering (CAiSE’2003).

Klagenfurt/Velden, Austria: 28-40.

Claypool, M., P. Le, et al. (2001). Implicit interest

indicators. Proceedings of the 6th international

conference on Intelligent user interfaces. Santa Fe,

New Mexico, United States, ACM: 33 - 40

Cruz, S. M. S. d., M. L. M. Campos, et al. (2004).

Monitoring e-business Web services usage through a

log based architecture. IEEE International Conference

on Web Services, IEEE: 61-69.

Herlocker, J. L., J. A. Konstan, et al. (2004). "Evaluating

collaborative filtering recommender systems." ACM

Trans. Inf. Syst. 22(1): 5-53.

Margaria, T. (2008). The Semantic Web Services

Challenge: Tackling Complexity at the Orchestration

Level. 13th IEEE International Conference on

Engineering of Complex Computer Systems, 2008.

ICECCS 2008, IEEE: 183-189.

Melville, P., R. J. Mooney, et al. (2002). Content-Boosted

Collaborative Filtering for Improved

Recommendations. Proceedings of the Eighteenth

National Conference on Artificial Intelligence (AAAI-

2002). Edmonton, Canada: 187-192.

OASIS. (1999, 1999). "ebXML." Retrieved 10

September 2008, from URL: http://www.ebxml.org/.

OASIS. (2004, October 2004). "UDDI Specification Draft

3.0.2." Retrieved 20 September 2008, from URL:

http://www.oasis-open.org/committees/uddi-

spec/doc/spec/v3/uddi-v3.0.2-20041019.htm.

Paolucci, M., T. Kawamura, et al. (2002). Semantic

Matching of Web Services Capabilities. Proceedings

of the First International Semantic Web Conference on

The Semantic Web, Springer-Verlag. 2342: 333-347.

Perugini, S., M. A. Gonçalves, et al. (2004).

"Recommender Systems Research: A Connection-

Centric Survey." J. Intell. Inf. Syst. 23(2): 107-143.

Pu, P., L. Chen, et al. (2008). "Evaluating product search

and recommender systems for E-commerce

environments." Electronic Commerce Research 8(1):

27.

Sapkota, B. (2005). Web Service Discovery in Distributed

and Heterogeneous Environment. Proceedings of the

WWW Service Composition with Semantic Web

Services (wscomps05) Workshop. Compiegne, France

Shen, Z. and J. Su (2007). On Complexity of The

Tightening Problem for Web Service Discovery. IEEE

International Conference on Service-Oriented

Computing and Applications, 2007. SOCA '07, IEEE:

29-38.

Thio, N. and S. Karunasekera (2005). Client profiling for

QoS-based Web service recommendation. Software

Engineering Conference, 2005. APSEC '05. 12th Asia-

Pacific.

Torres, R., S. M. McNee, et al. (2004). Enhancing digital

libraries with TechLens+. Proceedings of the 4th

ACM/IEEE-CS Joint Conference on Digital Libraries

Tuscon, AZ, USA, ACM, New York, NY: 228-236.

Trabelsi, S., J. C. Pazzaglia, et al. (2006). Secure Web

Service Discovery: Overcoming Challenges of

Ubiquitous Computing. Proceedings of the European

Conference on Web Services (ECOWS'2006).

Washington, DC, USA.

van der Aalst, W. M., H. A. Reijers, et al. (2007).

"Business process mining: An industrial application."

Inf. Syst. 35(5): 713-732.

W3C. (2004). "Web Services Architecture Requirements:

W3C Working Group Note 11 February 2004." 2010,

from http://www.w3.org/TR/wsa-reqs/.

Werthner, H., H. R. Hansen, et al. (2007). Recommender

Systems. Proceedings of the 40th Annual Hawaii

International Conference on System Sciences,

HICSS'07. Hawaii, IEEE Computer Society,

Washington, DC, USA: 167.

Yu, Q., A. Bouguettaya, et al. (2008). "Deploying and

Managing Web Services: Issues, Solutions, and

Directions." The VLDB Journal 17(3): 537-572.

i http://www.amazon.com/

ii http://www.lastfm.comq/

iii http://www.cs.indiana.edu/~sithakur/l542_p3/

iv http://www.grouplens.org/

WEBIST 2010 - 6th International Conference on Web Information Systems and Technologies

124