J. Software Engineering & Applications, 2010, 3, 1089-1106
doi:10.4236/jsea.2010.312128 Published Online December 2010 (http://www.SciRP.org/journal/jsea)
1089
Context-Based Recommendation Approach for
Measurement and Evaluation Projects
Hernan Molina1, Luis Olsina1, Gustavo Rossi2,3
1GIDIS Web, School of Engineering, National University of La Pampa, General Pico, Argentina; 2LIFIA, School of Informatics,
National University of La Plata , La Plata, Argentina; 3National Council of Scientific an d Te c hnical Research, La P lata , Argenti na
Received November 19th; 2010, revised November 30th, 2010; accepted December 10th, 2010.
ABSTRACT
Data, information and knowledge are recognized as useful assets for analysis, recommendation and decision making at
any business level of an organization. Providing the right information for decision making considering different user-
requirements, projects and situations is, however, a difficult issue. A frequently-neglected challenge is to cope with the
influence of contextual issues affecting the success of outcomes and decisions. Particularly, when conducting quality
evaluations in software organizations, it is of paramount importance to identify beforehand the contextual issues af-
fecting outco mes and interp retation s for measu rement an d evaluation projects. Therefore, the relevant context informa-
tion shou ld be clearly id entified, specified and recorded for performing more robust an alysis and recommendations. In
this work, a domain-independent context model and a mechanism to integrate it to any application domain is presented.
The context mod el is built upon a measurement and evaluation framework enabling quantification and semantic capa-
bilities. The context model is then integrated in the mentioned framework itself to enable recommendations in meas-
urement and evaluation projects.
Keywords: Context-Aware, Context Model, Recommender System, Measurement, Evaluation, C-INCAMI Framework
1. Introduction
It is well recognized that information and knowledge
play a central role in the success of organizations in-
volved in the development or use of software and web
applications. This is evidenced by the increasing amount
of research and events in areas related to information and
knowledge management, spanning a wide range of dif-
ferent fields such as e-business, e-learning, web enginee-
ring and Web 2.0, organizational memory management,
just to mention a few. In any case information cannot be
interpreted in isolation due to the influence of contex tual
factors such as temporal, functional, structural, and user
viewpoint aspects. Indeed, recent research has started
paying attention to the important role context plays on
how informatio n is handled and interp reted and how con-
text can be represented and gathered to that end. These
are the concerns of the so called context-aware approach,
firstly applied in mobile and ubiquitous applications [1]
where the need of context is evident as a key driver for
delivering content and services. However context-aware-
ness is also useful in other fields such as software and
web engineering. Regardless of the field of application, a
context model is a key issue and a starting point in suc-
cessfully applying a context-aware approach. To that aim,
we believe that in order to achieve an effective context-
aware application, the designed context model should
have support for context property types, both at a struc-
tural and semantic level, and for the contextualization of
domain concepts. Nevertheless we observe in existing
related works, regardless of the chosen modeling app-
roach, the lack of a sound approach towards the structur-
ing and quantification of context properties: context mo-
dels are built from scratch disregarding measurement
theory concepts, i.e. no meta-data about scales (and scale
types), units and quantification methods are specified. In
this sense, we argue that the quantification of context
properties is just the quantification of some entity's attri-
butes; thus, context models could benefit from existing
measurement models and frameworks to build upon th em.
In this work we present a context model built upon the
measurement and evaluation framework so called INCAMI
(Information Need, Concept Model, Attribute, Metric and
Indicator) [2]. This framework provides concepts such as
Entity, Attribute, Metric, Scale, Unit, among others,
which can be used as a foundation to build applications

Context-Based Recommendation Approach for Measurement and Evaluation Projects 1090
capable of specifying data a nd meta-data for measurement
and evaluation design and implementation. Another issue
to be addressed by a context-aware application is the
establishment of mechanisms to gather relevant context
information from the actual and relevant situation of the
entity of interest, and also to use that info rmation in pro-
viding context-based recommendations. Lastly, architect-
tural support to ease the integration of context-related
functionality with existing domain applications must be
provided in order to realize the benefits of having context
descriptions of the situation at hand. In this work a me-
chanism based on semantic catalogues was designed and
implemented together with a pr otot y pe of a cont e xt -bas ed
recomm ender system int ended to be use d by organizat ions
(details of this implementation are out of the scope of this
article).
In this article the contex t-aware approach is applied to
web engineering; particularly, we are i nterested in conte xt
inform atio n affect ing t he desi gn , im plem ent atio n and out-
comes interpretations for measurement and evaluation
(M&E) projects in software organizations. To this aim, it
is of param ount im portance to clearly identify, specify and
record from the start context information relevant to a
particular M&E project [3-6] to 1) assist in design deci-
sions—e.g. selecting a metric, whose correct applicability
depends on the target context, to quantify some entity
attribute-, and 2) to make explicit the underlying issues
that characterize a given entity or the project itself for
analysis and c omparison pur poses. In t he en d m ore r obust
analysis and recommendations could be achieved among
the whole organization’s projects. Sev eral proposals ex ist
in the literature covering different views for the M&E
domain—i.e. proces ses, conceptual m odels, methods [3-5,
7-10] although none of them cover the context issue in
such a comprehensive way to achieve the above men-
tioned capabilities.
Therefore we extend and improve our previous pro-
posal [11] in which a context model was included in the
INCAMI M& E framework all owing to specify in a clearly
structured way context-aware data and meta-data for
M&E projects as a foundation to support context sensitive
recommendations to M&E activities. The result was the
C-INCAMI (as for Contextual INCAMI) framework. The
extension presented here includes a mechanism to inte-
grate existing domain concepts in the context model and
to compare context descriptions as a foundation for a reco-
mmender system.
Summarizing, the main contributions of this article are
the following:
 A domain-independent context model built upon a
M&E framework—an enhancement of [11] and a
mechanism to integrate existing domain concepts
with the context model;
 A context comparison mechanism—based on simi-
larity and applicability metrics—allowing the reali-
zation of the above two context-aware recommen-
dation cases for M&E;
The result is an improved context-aware M&E frame-
work with recommendation capabilities that can be inte-
grated to existing application domains.
The organization for the reminder of this article is as
follows: Section 2 presents an overview of the applied
context-aware approach, followed by an explanation of
the proposed context model and the domain integration
mechanism. In Section 3 the application of the context-
aware approach to measurement and evaluation is pre-
sented. First, in subsection 3.1, we give details of how
context is specified and used; then, the design of metrics
for the context comparison mechanisms are presented;
lastly, a proof of concepts shows some details of the use of
the C-INCAMI framework and the supporting architect-
ture. Related work and discussi on are addressed in Section
4; and, fi nally, Secti on 5 discu sses, conclusions and future
work.
2. A Context-Aware Approach
The rationale of the context-aware approach is based on
the widely accepted premise that information cannot be
interpreted in isolation but rather considering its context.
Applying this approach implies knowing and processing
quantitatively the context attributes or properties that in-
fluence how domain information is interpreted and pro-
cessed, for which a properly designed context model is
necessary. A context-aware application is then one wh ich
depends on or uses context information to offer its ser-
vices [12].
Applying t he context-a ware approach implies decisions
on a set of issues, namely, a context model and mecha-
nisms to capture and use context information. Decisions
on these issues should be made according to the intended
purpose of the approach. These will decide on the general-
ity of the c ontext-a ware applicat ion, which would be desi-
rable to some extent, for the sake of reusab ility, but with-
out disregarding the trade-off with practical applicability.
But before engaging in any design or technical en-
deavor, it is necessary to understand what context is and
how it relates to specific-domain information. In [12-15]
authors provide slightly different definitions of context
with different purposes and target domains. For this mat-
ter, we agree with Dey’s general definition of context [12]:
“any information that can be used to characterize the
situation of an entity”, which may be “a person, place, or
object that is considered relevant to the interaction be-
tween a user and an application, including the user and
applications themselves”. However, rather than adopting a
definition of context as is, we analyzed the definitions

Context-Based Recommendation Approach for Measurement and Evaluation Projects 1091
found in the cited literature and observed some common
issues about context information:
 It is relative to a specific object or entity of interest;
 It should be relevant with regard to:
o The specific task related to that entity,
o The specific intention towards the entity related
to that task,
o The interaction of the entity with other entities
of the context regarding the task and purpose;
 It is composed of a set of relevant properties with an
associate d concept ualization (i.e. st ructure ) that m akes
their interpretation explicit. These properties may
belong to the entity of interest or relevant entities re-
lated to it.
Summarizing the above listing (see Figure 1), we con-
sider context information as the set of relevant properties
(and its values) of the relevant entities involved in the
situation. By situation we understand the summing of th e
entity of interest, the task and intention towards that entity
and the interaction between this and other entities; and, by
relevant, we mean t hat it may affect on how information is
interpreted regarding the situation at hand.
Regarding the context m odel a decision shou ld be made
about the modeling approach following alignment with
[16] taking into account the corresponding pros and cons
that should be analyzed for the particular needs of the
application. In any case, the model should enable repre-
sentation of context informati on and its integration to the
target domain based on the understanding of the charac-
teristics of context information, as listed before.
Regarding the mechanisms to capture or specify con-
text information two approaches arise [17], namely, con-
text can be specified and attached at some point in time
with the pertinent information to be used later as a key for
selection, or it can be updated dynamically to adjust the
system behavior actively. Since both approaches satisfy
different needs the choice must be taken considering the
intended purpose of the application.
Regarding the use of context information, mechanisms
to process data instances of the context model should be
designed. Basically, mechanisms are necessary to com-
pare two given context descriptions, one from the actual
Figure 1. Overview of the elements in the characterization
of context information.
situation or environment and the other from the relevant
entity in the application domain, or both context descrip-
tions from two separate application domain entities. These
mechanisms should produce a value that expresses in
some objective way the degree of similarity of the com-
pared contexts. Again, the concrete mechanism depends
on the purpose of the application.
The above considerations remain valid regardless of the
application domain. For instance, in the case of the ubi-
quitous computing field, where a mobile user is the focus
entity, context should be m odeled and mecha nisms should
be devised for instance to discover and capture relevant
informat ion f rom t he surro undi ng en vir onment of u sers i n
order to offer them better services (relevant context as-
pects in this case are time, location, device, activity, role,
etc.). In our proposal, like in [14,18,19], context is con-
sidered from the information and/or knowledge man-
agement standpoint , particu larly in a software engi neering
domain with the aim of improving interpretations and
analysis results. In this case, the entity of interest is any
information element involved in some task, and the rele-
vant context information comes from related entities in
the domain considered relevant with regard to the task and
purpose, e.g. related to processes, projects, resources,
stakeholders, etc.
In this work we focus on the application of the con-
text-aware approach to the measurement and evaluation
domain aimed at making more coherent measurement
designs, and more sound interpretations from evaluation
results among projects. Context descriptions will be spe-
cified both for actual situations of M&E projects and for
reusable information elements—for example metrics,
quality models. etc. —which may be applicable to specific
target contexts. These context sp ecifications will provide
ultimately the foundation for a M&E recommender sys-
tem.
2.1. Context Model
Based on the general characterizati on of context p resented
earlier a model to represent context information was de-
signed. To this aim, different representation approaches
were analyzed [16], taking into account a set of require-
ments of interest to our proposal—some of which are
common to the ub iquitous computing do main:
 context information should be able to be validated
against the context model both at structure and in-
stance level;
 contextual information should be precise and un-
ambiguous allowing the consistent reuse of it;
 the context model should be able to be integrated to
current systems;
 the model should be simple enough to be specified
and processed with the aim of maintaining the sys-

Context-Based Recommendation Approach for Measurement and Evaluation Projects 1092
tem cost-performance trade off;
 the model should be domain independent and allow
tailoring to cover any application domain;
An object-oriented modeling approach was chosen for
the high-level definition of the context model because of
its expressiveness and, at the same time, its understand-
ability for communication purposes. Moreover, because
of the requirement of domain independence and integra-
tion to existing domains, an ontology approach will be
used for the representation at the implementation level to
take advantage of its facilities to interconnect different
application do mains. Both approaches a re complem entary
because of their similarities.
In the above characterization we said that context is
composed of properties belonging to the entity of in terest
or relevant entities associated to it in a given situation. We
also argued that these context properties can be seen as
ordinary attributes of some en tity, thus the context mode l
could be nefit from existing m easurement fram eworks pro-
viding these concepts and those related with attributes
quantification. In this work we use the INCAMI M&E
framework [2] as a foundation to build our context model.
For that purpose we briefly present next only the neces-
sary framework's concepts; further details of the frame-
work will be shown in Section 3.
The INCAMI framework specifies the concepts and
relationships necessary to specify the design and imple-
mentation of measurement and evaluation activities.
However, in order to build the con text model only meas-
urement related concepts are used as presented in Figure
2. The gray boxes group related concepts as indicated by
their labels. The framework provides for the specification
of concret e Entities belo nging to s ome Entity Category. In
turn each category has a number of associated Attributes
that can be used to describe entities belonging to that cate-
gory. Each attribute can be quantified by a number of-
Metrics—although only one is used in a particular case.
Broadly speaking, each metric is defined by a Scale de
scribing the values the attribute m ay assume and a Method
indicating how that value can be obtained. Notice from the
related concepts (and their attributes) in the diagram of
Figure 2 that different scales and methods can be clearly
specified. To quantify a given attribute of some entity a
Figure 2. An UML class diagram showing some concepts and relationships of the INCAMI framework used by the context
model.

Context-Based Recommendation Approach for Measurement and Evaluation Projects 1093
Measurement is performed using the metric assigned to
the attribute. These measurements, and their correspond-
ing measured values, are then associated to the entity on
which they were performed.
In order to reuse the quantification capabilities of the
INCAMI framework and based on the characterization of
context information, the context model is obtained by
extending the concepts Entity and Attribute. The resulting
model (see Figure 3) includes the definitio ns of the con-
cepts and relationships necessary to specify context in-
formation and to contextualize (i.e. associate context
information to) domain-specific information. The model
was designed as a separate module to ease its integration
with existing systems.
The context model is composed of three concepts. The
main concept is Context which represents the relevant
state of the situation of an entity:
Definition 1 (Context) A special entity, described by a
set of context properties that characterizes the relevant
state of the situation of a given entity of interest. The
situation of an entity involves the task and purpose to-
wards that entity and its relationships with other entities
regarding that task and purpose.
We consider Context as a special kind of Entity in
which related relevant entities are involved (as depicted in
Figure 3 by the related Entity relationship). Relevant
entities are those whose attributes are relevant in the in-
terpretation of data relative to the task and purpose to-
wards the entity of interest.
The situation is described by attributes of the relevant
entities involved in the context of the entity of interest.
These attributes are Context Properties that, properly
specified and quantified, can be used to describe the re-
levant context of the entity of interest regarding the task
and purpose:
Definition 2 (Context Property) A attribute that de-
scribes the Context of a given entity; this attribute belongs
to (the definition of) one of the entities that participates in
the described Context.
A context property is described by the meta-data inhe-
Figure 3. The context information module.
rited from the Attribute class—i.e. name, definition and
objective—and also by the following meta-data, particu-
lar for its purpose:
 weight: the relative value of the context property
within a given context description; values range
between 0. 0 a n d 1.0.
 relevance: the pertin ence of using th e context prop-
erty to describe the context for a particular informa-
tion need or contextual entity.
 multiValueCriterion: logic criterion to determine if
multiple values will be allowed for the context
property and, if so, how they will be treated. Possible
values are AND, OR, XOR—the first two values
allow more than one value to be assigned to the
property, while the last allows only one (the use of
the logic properties of each operator are explained in
Section 3).
The first two class attributes a re m eant to be assigned a
value wh en the contex t pr op erty is being used to describe
a particular context instance because of its definition:
weight is relative to all the properties used to describe the
context, and its relevance depends on the case and ra-
tionale of why the prope rty i s rel evant. On th e other hand ,
the multi Value Criterion is intrinsic to t h e propert y so its
value is stored in its own definition.
As defined so far, this extension allows specifying the
context, for a giv en entity of interest, b y identifying, and
associating to it, related relevant entities and then adding
their attributes as context properties to the cont ext. Then,
metrics can be assigned to each c ontext prope rty and mea-
surements be performed on the corresponding entity;
these are then associated to the context object.
For the contextualization of domain-specific informa-
tion, the model also allows specifying the ideal context in
which a given context-sensitive entity could be applied or
used, namely the target context, based on the nature and
definition of that entity. These entities are called Con-
textual Entities:
Definition 3 (Contextual Entity) An entity whose
correct usage and/or interpretation is sensitive to the
context in which it is applied and/or analyzed.
As it is designed, the context model is pluggable and
reusable in existing applications, since the concepts in the
context model do not specify which entity situation it cha-
racterizes or which are the contextual en tities; rather, the
domain implementing the context-aware approach is res-
ponsible for deciding which concepts are to be contextu-
alized by establishing relationships with the corres pon-
ding elements in the context model. Thus, this model can
be used to describe context information for any applica-
tion domain, su ch as ubiquitous co mputing, even though
it was first conceived to apply to a web engineering envi-
ronment.

Context-Based Recommendation Approach for Measurement and Evaluation Projects 1094
Notice that the simplicity of th e context model d efined
by specializing measurement related concepts from ex-
isting frameworks. In this way details of how properties
are quantified and associated to the entities from where
they were taken remain in the extended measurement
framework, thus hidden from the context model.
2.2. Representation and Integration of Context
Information
As mentioned above an ontology approach will be used in
the representation of context information, taking advan-
tage of the characteristics of the approach and its simi-
larities with the object model used for th e context mode l.
Particularly, we decided to use RDF (Resource Descrip-
tion Framework) [20] considering its standard status and
since there is already a comprehensive set of supporting
tools for storage, management and validation. This way,
semantic catalogues could be implemented to store both
domain-specific as well as context data and meta-data. Of
course, this decision may cause a restriction to the im-
plementation of the ap proach in the sense that app lication
domains be e ncoded in RDF. Neve rtheless we believe t hat
currently RDF has become an industry standard for in-
formation interchange. In the worse case, a set of tools for
relational to RDF mapping does exist to fill the gap1.
Now, for the integration of the co ntext model with ap-
plication dom ains t wo aspect s need t o be ad dressed. Fi rst,
concepts in the ap plication domain can be contextu alized
by simply linking to Context descriptions or su b-classing
ContextualEntity. Second, the context model can be lin-
ked to the application domain, as shown in Figure 4, by
using the UR I (U nif orm Re source Identifiers) mechanism
of RDF. Thus, the attributes classURI, individualURI and
propertyURI in some of the concepts of the M&E model
(Figure 4(a)), such as Attribute, Entity and Entity Cate-
gory, are used to store the URI locator of the corre-
sponding definition in the application domain (Figure
4(b)). This way, as shown in Figure 4, each entity in-
volved in a context can be linked to the corresponding
entity description in the organization's databases and the
associated context properties can be linked to the corre-
sponding properties of those entities as well. Moreover,
the elements involved in the context model, entities and
context properties, could be automatically gathered and
specified from its source in the organization's databases.
3. Context-Aware Measurement and
Evaluation
Measurement and evaluation are key processes in achi-
eving quality for produ c ts and services offered by organi-
zations, particularly in software and web engineering,
where quality is a must for the success of any endeavor.
This is evidenced by the amount of researches and stan-
dards [3-6,8-10,21] that document the activities and arti-
facts needed to that aim. A common denominator among
these works is the assertion that the ex plicit specification
and storing of context descriptions is needed as a pa-
rameter for selecting metrics and interpreting evaluation
outcomes. However none of these works provide a thor-
ough proposal to perform these context-related activ ities.
In order to illustrate the above statement, consider the
following scenario in which we show how context in-
formation, if clearly structured, may help in a M&E de-
sign process. The scenario is based on an M&E case study
[22] we performed using INCAMI. Initially, the infor-
mation need is specified: the purpose is to improve the
external quality of an e-Commerce shopping cart, as
perceived by customer users. The shopping cart belongs
to the Cuspide.com WebApp, an Argentinian company
which sells books online. Then, non-functional require-
ments for the external quality are designed, including a set
of high-level characteristics and associated measurable
attributes. One of the attributes to measure is the Read-
ability of text in help and support pages, necessary to
determine the achieved satisfaction level for the Learn-
ability concept. During measurement design two metrics
are found in the organization’s catalogue to quantify this
attribute: the Gunning fog index2 and the Lexile text3 me-
trics, so one of them should be selected. According to the
available meta-data of each metric, that includes for in-
stance its objective, value interpretation, author, refer-
ences and scale-, the Lexile text metric’s values are in-
terpreted as a numeric representation of an individual’s
reading ability or a text’s readability (or difficulty), while
Gunning fog index’s values are interpreted as an indicator
of the number of years (of USA formal education) that a
person requires in order to easily understand the text on
the first reading. From these descriptions, either metric
could be successfully applied. Nevertheless, some in-
formation is missing in these descriptions since the Gun-
ning fog index metric is designed for English texts, while
the Lexile text metric supports both Eng lish and Spanish.
This information becomes important if we consider that
the Cuspide’s shopping cart conten t is written in Spanish.
Although this fact could have been included in the met-
ric’s interpretation, it would still need a manual inspection
to determine its applicability. Otherwise, by having a
clear and structured specification of the supported natural
language context property associated to each metric, this
choice can be made automatically. Thus, we argue that
measurement and evaluation is contextual, so explicit
context descriptions should be provided to support more
sound analysis and interpretations of projects outcomes.
1See developments of the W3C Workshop on RDF Access to Rela-
tional Databases at http://www.w3.org/2007/03/R d fRDB
2http://www.idph.state.ia.us/health_literacy/common/pdf/tools/gunning
p
df
3http://nces.ed.gov/pubs2001/200108.pdf

Context-Based Recommendation Approach for Measurement and Evaluation Projects 1095
(a) C-IMCAMI Object model (b) RDF graph from the organization’s data base
Figure 4. Example of the integration of context information with the application domain.
Next, we present an overview of the INCAMI framework
which will be extended to support this capability.
In Section 2.1, we presented an initial overview of
INCAMI, foc using onl y on m easurement rel ated conce pts.
Now we go further by giving more details about the
INCAMI framework. As mentioned earlier, INCAMI
provides a domain model defining all the concepts and
relationships needed to support M&E processes in soft-
ware and web projects, as part of an organization's quality
assurance program. Thus, data and meta-data about In-
formation Needs, Concept models, Attributes, Metrics,
Indicators, and related concepts (see Figure 5) can be
specified to document artifacts involved in the process.
INCAMI is based on an ontology [23] and also give
support to the WebQEM (Web Quality Evaluation Me-
thod) methodology [2]. In addition, a prototype tool and
an organizational catalogue were built [24] following the
INCAMI framework model to maintain the instances of
all defined concepts. So the catalogue can be used in
different M&E projects, providing reuse mechanisms and
making the check for meta-data consistency in analysis
processes easier. In turn, the INCAMI framework and the
defined M&E process can be used to instantiate methods
and tools.
The framework is structured in modules or packages,
namely: project management, non-functional require-
ments specification, measurement design and implement-
tation, and evaluation desig n an d im pleme nt ation (see [2]
for a thorough discussion of all modules):
 The project management module deals with the
specification of managerial information about M&E
projects;
 The non-functional requirements specification
module deals with the definition of the information
need and the speci fication of requirem ents by means
of one or more concept models (e.g. external quality
models, quality in use models, etc.) and their attrib-
utes;
 The measurement design and implementation mod-
ule deals with the specification of concrete entities to
be measured, the metrics selection (from the cata-
logue) to quantify the attributes, and the recording of
the gathered measures; this module is centered in the
metric concept;
 The evaluation design and implementation module
deals with the definition of indicators, decision cri-
teria and aggregation models that will help to enact
and interpret the selected concept model.
Briefly outlined, INCAMI follows a goal-oriented ap-
proach in which all the activities of the M&E process
(e.g. metrics and indicators selection, measurement and
evaluation execution, etc.) are guided by a stated Infor-
mation Need which specifies particular non-functional
requirements of some Entity for a particular purpose and
user viewpoint. Non-functional requirements are repre
sented by Concept Models including high-level concepts

Context-Based Recommendation Approach for Measurement and Evaluation Projects 1096
Figure 4. Diagram showing the main concepts of the INCAMI framework.
or characteristics, as in ISO 25010’s qu ality models [21],
as well as quantifiable Attributes associated with the en-
tity under analysis. Concept models are the backbone for
measurement and evaluation. Measurement is specified
and implemented by using Metrics, defining how to ob-
tain and represent attributes’ values; basically, a metric
represents meta-data about the measurement or calcula-
tion method and the measurement scale. Evaluation is
defined and implemented by using Indicators, which
specify how to interpret attributes’ values and high-level
concepts of the quality model. Also the framework is
organizati on centered in the sense th at the definition of the
involved co ncepts allows its integratio n and implementa-
tion along softwa re and web projects within an organi za-
tion, carrying out measurement and evaluation activities
as part of its quality assurance program.
In order to support the contextualization of these con-
cepts, the M&E framework must be extended using a
context-aware approach. In the following sub-sections we
present the application of the context model and related
mechanisms presented in the previous sections to the
INCAMI framework to obtain support for context-aware
M&E in a clear and structured way in software and web
projects. The improved framework is called C-INCAMI.
From the definitions presented in Section 2.1, we now
perform a thorough analysis of their elements in order to
apply them to the M&E domain based on the terms de-
fined in the INCAMI framework. It turns out that the
entity of interest, for which context is specified, is the
entity to be measured and evaluated to satisfy a specific
informat ion need. This c ould be any inst ance of a pr oduct,
resource, proc ess, or se rvice being ass essed. For exam ple,
in the above motivating scenario the entity of interest is
the Cuspide.com’s shopping cart. The situation from
which relevance is considered is the M&E of the entity of
interest to satisfy the stated information need purpose,
focus co ncept and associated non-functional requirements,
and viewpoint, and also the entities related to the entity of
interest, as far as they are relevant to the assessment
meaning those that could affect how the focus concept of
the evaluation is interpreted on the entity to be assessed.
The entities to assess are typically involved in a software

Context-Based Recommendation Approach for Measurement and Evaluation Projects 1097
project perf ormed by an or ganization. Therefo re, potential
relevant related entities from which information may be
extracted as context properties are:
 The project in which the entity is involved, including
the resources, artifacts, assets or other pr oducts used
in the project affecting the entity;
 The processes applied to the entity;
 The organization that runs the project itself;
 The stakeholders interested in the outcomes of the
evaluation;
 The environment external to that organization.
So context properties are obtained from properties as-
sociated to the definition of these concepts in the data ba se
schemata of the organization’s domain model. In turn,
values for these prop erties may b e obtained from proj ects
and organizational memories. For example, in the situa-
tion of t he “C uspide .com s ho ppi ng ca rt” evaluation in t he
motivating scenario, the entity “Cuspide.com WebApp”
where the shopping cart is in cluded—is relevant since its
attribute “supported natural language” may affect the
measurement and interpretation of readability of text in
the shopping cart h el p page a quality attribute included in
the non-functional requirements for external quality; th us
this attribute is considered as a context property. From th e
previous analysis, we could summarize a tailored defini-
tion of context as a special entity, described by a set of
context properties that characterize the relevant state of
the situation of an entity subject to M&E. The situation of
the entity is defined by the information need and its in-
teraction with other entities from which context proper-
ties are taken- regarding that need, as long as they affect
the implementation and interpretation of M&E for the
subject entity.
Once the context model has been understood and inter-
preted for the M&E domain it should be included in the
INCAMI framework as a new m odule. Then the necessary
relationships should be established to contextualize M&E
related concepts (see Figure 6). Two scenarios m ay be dis-
tinguished at this point. First, to describe the actual con-
text characterizing the current situation of the entity being
assessed in a given M&E project, the corresponding In-
formation Need is associated to a Context object (see the
characterizedBy association in the cited Figure). Second,
concepts in the INCAMI framework, whose definition
could determine they are more or less applicable to a
given target context are therefore considered a kind of
Contextual Entity. The identified contextual entities in
INCAMI are: Concept Model, Metric, and Indicator (see
the generalization relationships in the same Figure).
To complete the integration, all the context related data
and meta-data need to be stored i n a n organizational cata-
logue together with M&E data and meta-data to provide
all the projects with a sing le and consistent set of proper-
ties.
3.1. Specifying and Using Context
Now we discuss the specification and use of context in-
formation for the two scenarios of contextualization pre-
sented in the previous section.
In order to specify the actual context, which character-
izes the current situation defined by the information need
in a particular M&E project, the relevant entities involved
in the context are id entified. This is done by inspecting in
the organization’s catalogues the relationships of the en-
-tity to be assessed with entities belonging to different ca-
tegories such as product, process, project, resource, sta-
keholder and the hosting organization itself (or related
ones). Once they are identified, the relevant properties
associated to each of them should be selected together
with the corresponding value, again from the organiza-
tion’s catalogues. The choice of including each property
as relevant to describe the actual context is a decision of
the evaluator and its rationale should be specifi ed for each
context property. Actual context descriptions can be used
in the interpretation and analysis of M&E results. For ins-
tance, when comparing evaluations results from two
M&E projects assuming both of them are comparable on
the basis of the entity category, non-functional require-
Figure 5. Contextualization in C-INCAMI.

Context-Based Recommendation Approach for Measurement and Evaluation Projects 1098
ments, metrics and so forth the actual c ontext can be com -
pared to assure both projects are actually comparable;
otherwise, analysis results can be unreliable.
In the case of a target context, which characteriz es the
ideal situation in which a contextual entity of the INCAMI
framework is applicable, the specification should be done
a priori a nd in depe ndentl y of any parti cular M&E pr oject
and should be stored together with the contextual entity in
the organization’s catalogue. As before, the rational of
including each context property should be recorded for
each of them. In this case, no related entity is associated
with the target context. Target contexts can be used in
M&E design, when decisions may be required to select
context-sensitive artifacts for example, a metric to quan-
tify a given attribute to determine whether they can be
coherently applied to the M&E project by comparing it to
its actual context.
Now, whether describing actual or target context, the
relevant context properties need to be quantified in order
to characterize a given situation. When describing the
actual context, a Context object is associated with a
number of relevant entities and a set of Context Properties.
Since context properties are defined as an Attribute spe-
cialization (recall Figure 3) a Metric can be associated to
each property in the catalogue. This way, Measurements
can be performed on the relevant entity, involved in the
actual context, following the associated Metric specifica-
tion to quantify each context property. Conversely to
attribute quantification for evaluation purposes which
allows only one m easurem ent per at tribute, more th an one
measurement is allowed for each context property whose
multiValueCriterion value is either AND or OR. For in-
stance, in the introduced motivating scenario, for the
actual context description, the “supported natural lan-
guage” roperty is relevant and more than one value, i.e.
“panish” and “English”, can be recorded simultaneously
(see Figure 7)
If the multiValueCriterion is XOR then only one
measurement is allowed for the context property; this
would be the c ase of, for exam ple, the “developm ent team
size” property.
When describing the target context of a contextual en-
tity, since the target context describes th e ideal contex t in
which a given con textu al en tity is su itabl e—and sin ce we
sustain that the contextual entity can be applied success-
fully in more than one context—the multiple measure-
ments assigned to each property represent the diffe- rent
values that these different contexts may present for that
property. For instance, the “Lexile text” met r i c 3 to quan-
tify the help page Readability of Text attribute in the
Figure 6. Example of an actu al co n t ext sp ec ification.
3htt
p
://www.lexile.com/abou
t
-lexile/lexile-overview/

Context-Based Recommendation Approach for Measurement and Evaluation Projects 1099
Cuspide’s shopping cart is a contextual entity since its
applicability depends on the natural language in which the
text is written, being useful just for Eng lish OR Spanish
languages. Note that our context model uses to a great
extent the same INCAMI conceptual base for specifica-
tion and quan tification of properties (see Figure 8).
The two above scenarios are the base for the core func-
tionalities of the context-aware recommender system
which are detailed in the following sub-sections.
3.2. Context Based Recommendation System
So far the two scenarios of how context information is
used in C-INCAMI were addressed, stressing its role in a
context-aware M&E recommender system. Next we pre-
sent the calculations designed to compare contexts desc ri-
ptions to provide automated recommendations.
As commented before, contexts may be compared to
determine whether results from two M&E projects can
actually be compared or a contextual entity can be co-
herently applied in a given M&E project, e.g. when se-
lecting a metric at measurement design time. The com-
parison mechanism is based on metrics designed for the
two above scenarios: 1) the Contextual Similarity metric
intend to compare two actual contexts from two separate
M&E projects, and 2 ) the Contextual Entity Ap plicability
metric, intended to compare the target context of a con-
textual entity against an actual context. In both cases, the
metric values should be interpreted as the more simi-
lar/applicable when closer to 1. These metrics are dis-
cussed below.
3.2.1 Conte xtu al Simi l arity Metric
This metric applies when comparing two actual contexts
from two different projects. In this comparison, all the
properties specified in both contexts are considered alto-
gether regardless of whether or not they are present in
both contexts at the same time. Before going in to details,
we present the notati o n we use:
By AC we denote an actual context defined by
(CP,CV,CW) where:



1,,n
CP cpcp s the set of context properties
specified for a given context; they represent only
their meta-data (the properties definitions) as stored
in the organization’s catalogue;



11
,, n CVcp CVcpCVcpare the whole sets of
values describing some context, one set for each
context property, being

11
,, m CVcp CVCV
the set of values for the context property cpi ;



11
,, n CWcp CWcpCWcpis the set of relative
weights for each context property cpi; where each
CWcp i 

and
11
n
i
i=
CWcp=
.
So, given the actual contexts AC(a) = (CP(a), CV(a),
CW(a)), and AC(b) = (CP(b), CV(b), CW(b)) specified
for projects a and b respectively, we define:
Figure 7. Example of a target context specification.

Context-Based Recommendation Approach for Measurement and Evaluation Projects 1100
CP(a,b) = CP(a)CP(b) as the definitions of all the
distinct context properties specified by both AC(a) and
AC(b), and # CP(a,b) as the number of properties in
CP(a,b)
Then, contextual similarity CSim is calculated as the
weighted sum of the similarity between the equivalent
properties of AC(a) and AC(b) for all the properties in-
cluded in CP(a,b), normalized by #CP(a,b):
  













#
1#
CP a,bii
i
i=
CPWcp ,ACaCPsimcp ,ACa,ACb
CSimAC a ,AC b=,cpCP a,b
CP a,b
 (1)
Where CPS im is a metric used to obtain the similarity
between two equivalent properties i cp from both con-
texts AC(a) and AC(b) (presented below), and CPW is a
function that returns the weight of a given context prop-
erty cpk in the actual context of some entity e; if the
property is not present, it returns 0. In Equation 1 the
weight for each cpi are taken from the actual context
AC(a).
  


0otherwise
kkk
CPWcp ,ACe=CWcpecpCPe 




if
3.2.2 Context Property Similarity (CPSim) Metric
Remember that, from the definition of CP(a,b), i cp may
appear in one or both contexts, so the following is held:

 
i
j
ij kik
cpCP a,b ,
cpCPa|cp cpcpCPb|cp cp

 
As before, let

 
1,, km
CVcpCV CV   be the
values specified for context property k cp in the context
associated to the entity a and

k
CVcpb the values spe-
cified for the same context property in entity b, so we
define:




 bCVcpaCVcp=ba,CVcp kkk as all the distinct
values specifie d for context pr operty k cp from both
entities a and b altogether. Note that values in each
property comes from the same property definition,
as explained in Section 3.2;
 #CVcpk(a,b) as the number of val ues in CVcpk(a,b);




 bCVcpaCVcp=ba,ECVcp kkk

as the set of equal
values specified for context property k cp for both
entities; and,



#,
k
ECVcpa bas the number of values in


,
k
ECVcpa b.
Then, the similarity between two equivalent context
properties in the contexts of entities a and b is defined as:
  


  
#if
#
0otherwise
kkk
kk
ECVcp a,bcpCPacpCP b
CPSim cp ,ACa,ACbCVcpa,b





(2)
3.2.3. Contextual Entity Applicability Metric
The metric presented next can be used to determine
whether a given contextual entity can be applied coher-
ently to a given M&E project. The metric compares a
contextual entity’s target context with a project’s actual
context to determine the applicability degree of the con-
textual entity to the project. In this comparison, only the
properties specified for the target context are taken as
parameters, d ifferent from what occurs in the CSim metric
(Equation (1)) where all the properties specified for both
contexts where included. Now, the comparison is made on
the basis of the target context instead of the actual context,
which may specify properties not relevant for the con-
textual entity.
Therefore, given the actual context AC(p)=(CP(p),
CV(p), CW(p)) specified for project p and the target con-
text TC(e)=(CP(e), CV(e), CW(p)) specified for the con-
textual entity e, the contextual applicability CApp is cal-
culated as the weighted sum of the app licability between
the equivalent properties of TC(e) and AC(p) for all the
properties included in CP(e), normalized by #CP(e):

 









#
1#
CP eii
i
i=
CPWcp,TCeCPAppcp ,TCe,ACp
CApp =,cpCPe
CP e
 (3)
Here, the weight returned by CPW for each cpi are
taken from the target c ont e xt TC(e).
3.2.4. Context Property Applicability (CPApp) Metric
Unlike the CPSim metric (Equation 2), in order to perform
the calculation, the CPApp metric takes into account the
multiValueCriterion of the context property in the target
context. When in a target context’s property the OR
criterion is specified, it should be checked that at least one
value of the target context’s property be present in its
actual context’s counterpart. If the AND criterion is spe-

Context-Based Recommendation Approach for Measurement and Evaluation Projects 1101
cified, it should be checked that all values in the target
context’s property are present in its actual context’s coun-
terpart. So, CPApp is calculated as follows:
If   bCPcpaCPcp kk  does not hold , the n

 0=pAC,eTC,cpCPApp k
Otherwise, if the multiValueCriterion is OR or XOR
then



 pAC,eTC,cpCPApp k = Equation (2)
if the multiValueCriterion of k
cp is AND then
 



1if# #
0otherwise
kk
k
ECVcpe, p=CVcpe, p
CPAppcp ,TCe,ACp




It is worth mentioning that for the above discussed
metrics, the amount of context properties available in each
compared context is an important issue. If relevant prop-
erties are omitted when specifying either contexts in-
volved in the compariso n, the final results for the degr ee
of similarity/applicability will be undervalued. This may
lead to unreliable results and consequently to wrong in-
terpretations and decisions. That is why special attention
in the selection of the proper relevant context pro perties in
both actual and target contexts should be paid.
3.3 Proof of Concept
By the following proof of concept, which is based on a
case study shown in [22], we try to illustrate how the
context-aware approach in the C-INCAMI framework
works. The starting point is to define non-functional re-
quirements, and particularly the Information Need as
shown in Table 1.
Next, the actual context characterizing the information
need of this project should be specified. To this aim, the
relevant entities related to the “ Cuspide’s shopping cart”
are identified by inspecting the organization’s domain
catalogues for related products, processes, projects, re-
sources and stakeholders. As a result, we found that the
only related relevant entity is the “Cuspide e-Commerce
WebApp” product, which contains the entity to assess,
because some of its properties may affect the way meas-
urement and/or interpretation is performed on the external
quality of the shopping cart.
The next step is to select from the catalogue the rele-
vant properties associated with the above entity and to
quantify them properly. Table 2 presents the properties
available for the actual context. Each context property
contains all meta-data related to its definition (recall
Figure 3) of which only name, definition and relevance
are shown, and the assigned value describing the actual
state of the relevant context for that property.
After the context specification is completed a concept
model is selected and/or designed according to the par-
ticular evaluation need s. For this case an external quality
model was taken from the M&E catalogue and modified
to meet the project's needs as shown in Figure 9. Then the
measurement design activity should be performed so a
metric is selected and assigned for each attribute in the
external quality model. In the example, to quantify the
attribute Readability of text (coded 1.2.1.2 in Figure 9),
two metrics are found in the catalogue (see Table 3) to-
gether with the context where they are applicable, i.e.
their target contexts.
As discussed before, the applicability of both metrics
depends on the natural language in which the content is
written, as specified by the “Supported Natural Lan-
guage” property in their target contexts. From these spe-
cifications we can see the “Gunning fog index” can only
be used for English texts while the “Lexile text” metric
can be used either for English or Spanish texts. Therefore,
their applicability to quantify the “Readability of text”
External Quality
1 Usability
1.1 Understandability
1.1.1 Icon/label ease to be recognized
1.1.2 Information grouping cohesiveness
1.2 Learnability
1.2.1 Help and Support
1.2.1.1 Help page
1.2.1.2 Readability of text
1.2.2 Predictive information for link/icon ...
1.3 Operability
1.3.1 Control permanence
1.3.2 Expected behavior of controls
2 Information Quality
2.1 Content Suitability
...
Figure 9. Excerpt of the external quality model for the cus-
pide’s shopping car t [19].
Table 1. The information need for the cuspide. com shopping cart case study.
Purpose Focus Point of view Entity Category Entity
Improve External Quality Customer user WebApp’s shopping cart Cuspide WebApp’s shopping cart

Context-Based Recommendation Approach for Measurement and Evaluation Projects 1102
Table 2. The properties of the actual context characterizing the information need.
Name Market Scope
Definition The scope of countries to which the marketing of the e-Commerce Web Site’s products are constraint.
Relevance May affect outcomes interpretation regarding Content Sui tability, particularly Coverage. Related to the Web-
site owner’s coun try property
Scale/Unit [domestic, regional, global] (nominal categorical scale)
Value global
Name Website owner's country
Definition The country in which the Web site's owner company resides.
Relevance May affect outcomes interpretation regarding Content Sui tability, particularly Coverage. Related to Market
scope property
Scale/Unit An enumeration of the countries' names (nominal categorical scale).
Value Argentina
Name Supported natural language
Definition Natural languages in which the content of the Web site is written.
Relevance M ay help in measurement design since the Web site's text is part of the content whose quality will be measured
and evaluated
Scale/Unit An enumeration of the natural languages' names (nominal categorical scale).
Value Spanish
Name Software release life-cycle stage
Definition Stage that describes the stability of a released p iece of software as the development process proceeds.
Relevance May affect outcomes interpretation since different quality levels may be observed at different stages.
Scale/Unit [Pre-alpha, Alpha, Beta, Release Candidate, Release to Marketin g, General Availability, Live Release] (ordinal
categorical scale).
Value Live Release
Name Content provider role type
Definition Role or User type that provides the main content of the Web s i t e .
Relevance May affect outcomes interpretation since different quality levels may be observed on content provided by
different types of providers .
Scale/Unit [owner, registered user] (nomin al categorical scale).
Value owner
Table 3. Metrics to quantify the Readability of Text attribute and their target contexts.
Metric Name Lexile Text Numerical Scale
Objective To obtain a numeric representation of a text’s readability.Value type INTEGER
Value Interpre-
tation
Number representing an individual’s reading ability or a
text’s readability. The lower the value, the easier to read.
Scale type ABSOLUTE
References http://nces.ed.gov/pubs2001/200108.pdf Representation DISCRETE
Author MetaMetrics, Inc. Unit Lexile
Target Context
Property name Entity Value Logic Operator
Supported Natural Language Cuspide.com WebApp Spanish, English OR
Metric Name Gunning Fog Index Numerical Scale
Objective To obtain the understandability of a piece of text. Value type INTEGER
Value Inter-
pretation Number of years of U.S. formal education that a
person requires to easily understand the text on the
first reading. The lower the value, the easier to
understand.
Scale type ORDINAL
References http://www.idph.state.ia.us/health_literacy/commo
n/pdf/tools/gunning.pdf Representation DISCRETE
Author Robert Gunning Unit Years
Target Context
Property name Entity Value Logic Operator
Supported Natural Language Cuspide.com WebAppEnglish OR
attribute in the situation of the M&E project can be de-
cided by comparing their corresponding target contexts
against the actual context characterizing the M&E pro-
ject’s information need.
This comparison can be achieved automatically by
computing the applicability degree of each metric’s target
context to th e actual context, u sing the CApp metric (see
Equation (3)). Then, evaluators will be able to select the

Context-Based Recommendation Approach for Measurement and Evaluation Projects 1103
more applicable one, by observing th e applicability v alue
of each metric. These calculations are illustrated next.
We first calculate the CApp metric on the “Lexile text”
metric. We begin by calculating the CPApp metric for its
only context property: “Supported natural language” So
let p be th e M&E project on which the “Cuspide’s shop-
ping cart” is assessed and lexile be the contextual entity;
then:


Supported natural language{''''}CVlexile=english , spanish


 }''{
language natural Supportedspanish=pCV



Supported natural language{'',''}CVlexile,p=englishspanish


 2# language natural Supported=plexile,CV


 }''{
language natural Supportedspanish=plexile,ECV
  1,# language natural Supported=plexileECV
So, according to the specification of the CPApp metric
and applying Equation (2), results:
 


#0.5
#
Supportednaturallanguage
Supportednaturallanguage
ECVlexile, p
CPApp Supportednaturallanguage,TC lexile ,ACp==
CVlexile,p
And, since “Supported natural language” is the only
property specified in the context of the lexile metric, by
calculating Equation (3) we obtain: CApp(lexile, p) = 0.5.
Besides, by performing the above calculi on the Gunning
fog index metric it results that: CApp(gunning,p) = 0. As
we can see from these outcomes, the “Lexile text” metric
got the highest degree of applicability, so it is chosen to
quantify the “Readability of text” attribute.
Note that the above calculations could appear a bit
comp lex to ma ke j ust a s impl e co mpari son . Nev erth eles s
if more context properties woul d be specifie d in the target
context of the contextual entity, it would require only to
calculate the CPApp metric for each additional context
property and then compute CApp as defined in Equation.
3. When a large amount of attributes and their potential
metrics are involved, the advantage of using this proce-
dure is paramount, even more if the procedure can be
automated in a recommender system.
4. Discussion and Related Work
In this article we argue that being able to specify in an
explicit and formal way the relevant context of M&E
projects, is of paramount importance. Thus we have ap-
plied the context-aware approach to a M&E framework
enabling the usage of context information to recommend,
at design time, the better contextual entity in a given
project or to perform, at later stages, more robust com-
parative analysis among projects. Disregarding this in-
formation may lead to less reliable analysis and recom-
mendations [25].
Regarding the applied context-aware approach, it is
worth mentioning that although first designed for ubiq-
uitous computing, it can be successfully applied to any
domain if a thorough analysis of the relevant entities types
and their properties is performed. To this aim, a model
should be designed to represent both context properties
data (values) and metadata, defining in an unambiguous
way their meani ng and rep resenta tion rules (e.g. measu re-
ment methods, scales and units). The chosen model and
representation should match the intended purpo se for the
context information taking into account the trade-off
between acceptable complexity and needed expressive-
ness.
Several works can be found in the literature app lying a
context-aware approach. For instance, in [18] a process to
identify context descriptions from software engineering
research papers and a method to compare them, using a
hierarchical clustering algorithm , are presented. The com -
parison outcome is then used as an affinity parameter to
compare the results presented in the papers. In that work,
context information is recorded in a single-entry preset
template with no other definition than the entry name
making it more difficult to use it in intepretation activities.
In [19], a context-aware approach to represent and inte-
grate static enterprise information and dynamic process
information in knowledge management is presented. Con-
versely to the previous work, this approach includes a
generic cont ext kn owl edge st ruct ure model based o n a set
of meta-information elements at different conceptual
levels for context description that can be interconnected
with ontologies in the business environment. The ap-
proach proposes to use the foundations of the Semantic
Web to store and interconnect context descriptions and
enterprise knowledge altho ugh nothing is said abou t how
to use or p rocess context informati on. In [14] the context-
aware approach is applied in a web engineering process
for the adaptation of applications to deliver context-rele-
vant services and/or information to end users on a dis-
covery basis. The proposal includes a conceptual model
based on context parameters that are connected with a
certain weight, i.e. a degree of relevance, to elements of
an existing domain ontology. A preset categorization of
context parameters (User&Role , Pr ocess &Task, Location,
Time, Device) to satisfy a broad variety of web applica-
tion scenarios is provided, each one requiring specific
values relevant to the application domain. Authors argue
that such a categorization allows for a large degree of r e-
use when usi ng conte xt catalogues. Neverthe less this app-

Context-Based Recommendation Approach for Measurement and Evaluation Projects 1104
roach may not be the most ap propriated for dom ains othe r
than ubiquitous and pervasive computing (such as soft-
ware engi neering as in our work). For the representational
aspect authors chose OWL (Ontology Web Language).
The proposal also describes a scenario of use of how con-
text is used in the engineering process of developing an
application. Authors in [26] present a version model that
supports context-dependent alternatives and a matching
algorithm for context-aware queries to provide context-
aware data managem ent. These elements are im plemented
in an object-oriented database system with which a mo-
bile tourist information system was implemented. The
model, that supports typing of context properties, was
built using an object-oriented data model and the imple-
mentation is based on a generic context engine both of
their own. T he proposed mode l allows linking any context
property to any application domain concept. A common
issue of these works, and others, is the lack of a sound
conceptual base when it comes to context properties quan-
tification, i.e. few or no metadata is specified about the
scales (and scales types) and units in which values are
expressed nor the methods used to obtain them. Conse-
quently, interpretations and comparisons may lead to in-
valid conclusions.
In this work, having analyzed previously different re-
presentation approaches [16] based on a set of require-
ments particular to our purpose, we designed a context
information model using a measurement and evaluation
framework as a foundation. This way, capabilities related
to properties quantification are reused resulting in a
clearer and simpler definition of the concepts related to
context specification. The designed context model can be
implemented as a separate module to ease its integration
to existing systems. Also, its representation in RDF al-
lows connecting its elements with existing application
domains and to implement context catalogues to enable a
more consistent reuse of context properties definitions.
Also a mechanism to compare context descriptions was
designed, although keeping a simple matching approach.
The resulting context-aware approach has the following
characteristics:
 Simple: a context property is defined as a speciali-
zation of an attribute which is alr eady def ined along
with measurement capabilities in a well structured
framework;
 Flexible: the context model can be plugged to any
application domain and reuse its entities and attrib-
utes definitions to specify con text descriptions;
 Semantically defined and able to be validated: the
semantic and structure of each context property is
explicitly defined in the organization's data base
schemata from which the corresponding instances
were gathered ;
Regarding the application of the context-aware ap-
proach to the M&E, the need of being aware of the rele-
vant context i n this dom ain has bee n previously expressed
in several works although none of them have provided a
solution to explicitly specify an d use context information
as we propose. For example, the ISO-15939 [5] measure-
ment process states that “Characteristics of the organiza-
tional unit that are relevant to selecting measures and
interpreting the information products shall be explicitly
described” providing “the context for measurement” in
order to m ake cl ear “the as sum ptions t hat it embodi es an d
constraints that it imposes”. In addition it recommends
that context should be stored when collecting data to
verify it, understand it, and evaluate it. However, no
formal specification is represented to deal with such a
characterization. In [6], authors remark the importance of
having explicit representations of context to be effective
in interpreting and acting on measurement resu lts. In this
sense they argue that “every process operates in an envi-
ronment that contributes to or detracts from its prospects
for success”, and enum erate a number of factors which, if
quantified, may help to interpret data gathered from the
entities under examination. Guidelines are provided to
specify context information and to d etermine its bounda-
ries in terms of the entities (and their relevant attributes)
affecting the entity bein g measured and analyzed. Exam-
ples of e nviro nmen t descri ptions are gi ven a lthou gh us ing
a simple mechanism, i.e. a template form. In [4] authors
present a GQM-based (Goal Question Metric) process
model for measurement programs, which starts by char-
acterizing "the current project and its environment with
respect to existing models and metrics". One of th e main
principles in this work is that measurement goals should
be explicit, including the environment of the analysis
involving “the organization, the project, process model
used”, among others. Also a “process prestudy” is speci-
fied in which existing measurement documentation from
“programs performed in the same or a comparable con-
text” is used, yet no specificatio n of how to perform this
comparison is presented. Other authors as Kitch enham et
al. [10], an d Bri and et al. [3] advocate for t he explicitness
and use of context information b ut a structured pr oposal to
achieve such a goal is missing.
Ultimately, the application of the context-aware ap-
proach to the INCAMI framework, as presented in this
article represents a contribution in the sense that a clear
and structured support to the specification of context
information for M&E projects in software organizations
has often been ne gl ect ed.
5. Conclusion and Future Work
In this article we have discussed the application of the
context-aware approach for recommendation capabilities

Context-Based Recommendation Approach for Measurement and Evaluation Projects 1105
in the INCAMI M&E framewor k result in the C-INCAMI
framework. We have noted, however, that the designed
context information representation can be used in any
domain so in this sense it can be considered a domain-
independent approach.
In addition, we have shown, as an important contribu-
tion, that the a pplicat ion of t he cont ext-a war e appr oach t o
the M&E domain clearly benefits two scenarios: 1) by
assisting in the recommendation of design solutions, and 2)
by maki ng the inte rpretati on of out comes am ong diffe rent
evaluation projects more consistent and comparable. Par-
ticularly, we have illustrated as proof of concep t the case
where the more suitable metric (the contextual entity) was
recommended at measurement design time for the Cus-
pide’s shopping cart evaluation project (the actual con-
text).
In both cases the foundation for the M&E recom-
mender system is the proposed context specifications. As
another contribution we have provided the rationale of
how to quantify the context properties -reusing the con-
cept of Attribute and Metric from the INCAMI frame-
work- as well as the specification of the contextual simi-
larity and applicability metrics that con stitute the core of
the recommender system.
As result, C-INCAMI currently offers not only a simple,
flexible, goal-oriented organization approach but also a
consistent, semantically defined and able to be validated
mechanism for representing context information, there-
fore, giving support in the design and selection of con-
textual entities as well as in the consistent interpretation
and analysis of results coming from similar M&E pro-
jects.
Future work will focus on improving the matching ap-
proach to context comparison. Lastly, an in-progress line
of research is the implementation of an ontology-based
organizational memory application and its integration
with the contextual C-INCAMI framework and tool. That
is, the knowledge acquired and stored in a case knowledge
base from lessons learned in M&E projects may serve as
the basis for a recommendation system to support more
robust decision making.
6. Acknowledgements
This work and line of research is supported by the
PAE-PICT 2188 project at UNLPam, from the Science
and Technology Agency, Argentina.
REFERENCES
[1] M. Baldauf, S. Dustdar and F. Rosenberg, “A Survey on
Context-Aware Systems,” International Journal of Ad
Hoc and Ubiquitous Computing, Vol. 2, No. 4, 2007, pp.
263-277. doi:10.1504/IJAHUC.2007.014070
[2] L. Olsina, H. Molina and F. Papa, “How to Measure and
Evaluate Web Applications in a Consistent Way,” Web
Engineering: Modelling and Implementing Web Applica-
tions, G. Rossi, O. Pastor, D. Schwabe and L. Olsina,
Editors, Springer, London, 2008, pp. 385-420. doi:
10.1007/978-1- 84628-92 3-1_13.
[3] L. C. Briand, S. Morasca and V. R. Basili, “An Opera-
tional Process for Goal-Driven Definition of Measures,”
IEEE Transactions on Software Engineering, Vol. 28, No.
12, 2002, pp. 1106-1125. doi:10.1109/TSE.2002.1158285
[4] C. Gresse, B. Hoisl and J. Wüst, “A Process Model for
QGM-Based Measurement,” Gresse et al., 1995.
[5] ISO/IEC: ISO/IEC 15939: 2002—Software Engineer-
ing—Software Measurement Process, 2002.
[6] R. E. Park, W. B. Goethert and W. A. Florac, “Goal-
Driven Software Measurement: A Guidebook,” Software
Engineering Institute, Carnegie Mellon University Pitts-
burgh, Pittsburgh, 15213, 1996.
[7] A. Abran, A. Sellami and W. Suryn, “Metrology, Meas-
urement and Metrics,” 9th IEEE International Software
Metrics Symposium (METRICS 2003), Sydney, Septem-
ber 2003, 2003, pp. 2-11.
[8] CMMI Product Team, “CMMI for Development, Version
1.2,” Software Engineering Institute, Carnegie Mellon
University Pittsburgh, Pittsburgh, 15213, 2006.
[9] ISO/IEC: ISO/IEC 14598-1: 1999—Information Tech-
nology—Software Product Evaluation—Part 1: General
Overview, 1999.
[10] B. A. Kitchenham, R. T. Hughes and S. G. Linkman,
“Modeling Software Measurement Data,” IEEE Transac-
tions on Software Engineering, Vol. 27, No. 9, 2001, pp.
788-804. doi:10.1109/32.950316
[11] H. Molina and L. Olsina, “Assessing Web Applications
Consistently: A Context Information Approach Web En-
gineering,” 2008 ICWE’08. Eighth International Confer-
ence on Web Engineering, 2008, pp. 224-230. doi:
10.1109/ICWE.2008.21
[12] A. K. Dey, “Understanding and Using Context,” Personal
and Ubiquitous Computing, Vol. 5, No. 1, 2001, pp. 4-7.
doi:10.1007/s007790170019
[13] L. Gong, “Contextual Modeling and Applications,” IEEE
International Conference on Systems, Man and Cyber-
netics, 2005, pp. 381-386. doi:10.1109/ICSMC.2005.
1571176
[14] J. W. Kaltz, J. Ziegler and S. Lohmann, “Context-Aware
Web Engineering: Modeling and Applications,” RIA Re-
vue d’Intelligence Artificielle, Special Issue on Applying
Context Management, Vol. 19, No. 3, 2005, pp. 439-458.
[15] T. Strang, C. Linnhoff-Popien and K. Frank, “COOL: A
Context Ontology Language to enable Contextual Inter-
operability,” Proceedings of 4th IFIP WG 6.1 Interna-
tional Conference on Distributed Applications and Inter-
operable Systems (DAIS2003), 2003, pp. 236-247.
[16] T. Strang and C. Linnhoff-Popien, “A Context Modeling
Survey,” Workshop on Advanced Context Modelling,
Reasoning and Management, UbiComp 2004—6th In-

Context-Based Recommendation Approach for Measurement and Evaluation Projects 1106
ternational Conference on Ubiquitous Computing, Not-
tingham, 2004, pp. 34-41.
[17] P. Dourish, “What We Talk about When We Talk about
Context,” Personal and Ubiquitous Computing, Vol. 8,
No. 1, 2004, pp. 19-30. doi:10.1007/s00779-003-0253-8
[18] D. Cruzes, M. G. Mendonça, V. Basili, F. Shull and M.
Jino, “Using Context Distance Measurement to Analyze
Results across Studies,” ESEM ’07: Proceedings of the
First International Symposium on Empirical Software
Engineering and Measurement, IEEE Computer Society,
Washington DC, 2007, pp. 235-244.
[19] W. Huang and T. Tao, “Adding Context-Awareness to
Knowledge Management in Modern Enterprises,” Second
IEEE International Conference On Intelligent Systems,
2004, pp. 393-398
[20] “Resource Description Framework Standards,” The
World Wide Web Consortium, http://www.w3.org/ stan-
dards/ techs/ rdf#stds
[21] ISO/IEC FCD 25010.3 Software engineering—Software
product Quality Requirements and Evaluation (SQuaRE)
—Quality Model and Guide, 2009.
[22] L. Olsina, G. Rossi, A. Garrido, D. Distante and G. Can-
fora, “Web Applications Refactoring and Evaluation: A
Quality-Oriented Improvement Approach,” Journal of
Web Engineering, Vol. 7, No. 4, 2008, pp. 258-280.
[23] M. Martin and L. Olsina, “Towards an Ontology for
Software Metrics and Indicators as the Foundation for a
Cataloging Web System,” Proceedings of IEEE Com-
puter Society (1st Latin American Web Congress), Santi-
ago de Chile, 2003, pp. 103-113.
[24] H. Molina, F. Papa, M. Martin and L. Olsina, “Semantic
Capabilities for the Metrics and Indicators Cataloging
Web System,” Engineering Advanced Web Applications,
Rinton Press Inc., 2004, pp. 97-109.
[25] B. Kitchenham, D. R. Jeffery and C. Connaughton,
“Misleading Metrics and Unsound Analyses,” IEEE
Software, Vol. 24, No. 2, 2007, pp. 73-78. doi:10.1109/
MS.2007.49
[26] M. Grossniklaus, M. C Norrie, “An Object-Oriented Ver-
sion Model for Context-Aware Data Management,” Pro-
ceedings of WISE 2007, 8th International Conference on
Web Information Systems Engineering, Nancy, France,
December 2007, pp. 398-409.