=Paper=
{{Paper
|id=None
|storemode=property
|title=Relations Extraction on Patterns Lacking of Resulting Context
|pdfUrl=https://ceur-ws.org/Vol-867/Paper30.pdf
|volume=Vol-867
|dblpUrl=https://dblp.org/rec/conf/icwit/HachemiA12
}}
==Relations Extraction on Patterns Lacking of Resulting Context==
https://ceur-ws.org/Vol-867/Paper30.pdf
Relations extraction on patterns lacking of
Resulting Context
Asma HACHEMI1, Mohamed AHMED-NACER1
1
Computer science department, USTHB, Algiers, Algeria
ashachemi@usthb.dz, anacer@mail.cerist.dz
Abstract. Many software patterns are available nowadays. They allow the re-
use of proved solutions in various areas of software engineering, but are ex-
pressed in different formalisms. This diversity is detrimental for patterns re-
use, since it is difficult to compare and compose heterogeneous patterns (pat-
terns expressed in different formalisms). Moreover, patterns composition is
based on inter-patterns relationships, that are difficult to discern if they are not
explicit. Thus, an automatic method that extract non explicit relations between
patterns even if those latter are heterogeneous, becomes a necessity. In this
context, we improve an existing method of automatic inter-patterns relations
analysis. As many patterns lack of resulting context, our aim is to enable that
method to extract relations on this kind of patterns.
Keywords : Patters formalisms, inter-patterns relationships, automatic rela-
tions extraction.
1 Introduction
Nowadays, the WWW supplies an increasing amount of knowledge covering di-
verse domains. Among others, it supplies a large number of software patterns that are
a formidable tool allowing the reuse of proved solutions, in various areas of software
engineering. A software pattern presents an issue to a recurrent problem, by offering a
proven solution. Software patterns need to be composed, in order to solve complex
problems that are not dealt by a single pattern. Inter-patterns relationships are on the
basis of the patterns composition. However, it is difficult to discern these relations if
they are not explicit in each pattern. Moreover, even if those relations are explicit,
they are limited to intra-catalog relationships.
Indeed, a pattern is expressed through a pattern formalism, which is a syntactic
structure of the pattern content. The majority of pattern formalisms in the literature
differ in the number and degree of detail of their items. So, it is difficult to interpret
and compare heterogeneous patterns. It is also difficult to compose them into larger
solutions; a fact which is detrimental for patterns reuse.
This paper presents our approach that improves an existing method of automatic
inter-patterns relations analysis. This method is the first automatic approach which
handles relations between heterogeneous patterns, cross different catalogs. The aim
of our improvement is to enable that method to handle more patterns formalisms;
specially, those formalisms lacking of resulting context. So, works related to inter-
patterns relationships extraction are described in section 2. Our approach for automa-
tic relations analysis on patterns lacking of resulting context is presented in section 3.
Finally, we conclude this paper and give some research perspectives in section 4.
282
2 Related works
Many researcher were interested in defining inter-patterns relationships, like [1],
[2], [3], [4], … but very few works treat relationships extraction. In this area, Prab-
hakar et al. [15] propose a graphical model called Design Decision Topology Model,
in order to represent design patterns and extract relationships between them; unfortu-
nately, this work is limited to the analysis of relations on design patterns only. The
method of Kubo et al. [10] is the first automatic approach able to extract relations on
heterogeneous patterns, belonging to different catalogs, and is the only approach able
to deal with different kinds of software patterns. Kubo et al. method is an interesting
approach based on its own pattern model (consisting of Starting Context, Forces,
Resulting Context), and on several text processing techniques (stop word removal
[11], stemming [12], the TFIDF term weighting [13], vector space model [11] and the
cosine similarity). However, Kubo et al. method is not able to treat patterns which
lack of Resulting Context, so we propose to improve it to extract relations on this kind
of patterns.
4 Our approach
Our approach towards an automatic way to analyze relations between patterns is
based on Kubo et al. method, that we propose to improve in terms of the pattern forms
handled. As many patterns do not express the Resulting Context explicitly (like those
in [5], [7], [8], [9], [14], …), they cannot be represented and analyzed by the mo-del
of Kubo et al.. Therefore, a value-added of our approach is the proposition of a solu-
tion to represent this kind of patterns and analyze relations on them.
Resulting Context is the result or product generated by the pattern application. So,
each pattern has its resulting context either explicit in a dedicated section, or implicit-
ly given in the Solution section. Here is an example of a resulting context expressed
within the Solution : The pattern Declare Before First Use [8] aims to ensure that the
declaration of an element is positioned before the reference to it (in an XML docu-
ment). The resulting context of this pattern is the increase of the probability of treat-
ing the document in a single pass. Actually, this resulting context is expressed within
the Solution section :“This gives the processing software a better chance of doing a
single pass traversal of the document”.
Our idea is to overcome the absence of a dedicated section for Resulting Context
by using the Solution section, so that one be able to represent the third element of the
pattern model (Resulting Context). A such use of the Solution does not alter the sig-
nificance of the relationships that we are interested in (Starting-Starting [10],
Same[10], Resulting-Starting[10], Uses[6] and Refines[6]). The reasons are :
The relation Starting-Starting : The analysis of this relation is based on the
Starting Context [10]. The Resulting Context is used only to represent the
pattern in the Kubo et al. model. Thus, the use of the Solution section instead
of the Resulting Context does not affect the meaning of this relation.
283
The relations Refines : The analysis of this relation is based on the Starting
Context and Forces [6]. The Resulting Context is used only to represent the
pattern in the Kubo et al. model. So, the use of the Solution instead of the
Resulting Context does not alter the meaning of this relationship.
The relation Same : When two patterns share the same Starting Context and
the same Solution, this means that these two patterns deal with the same
problem and provide the same result. Thus, we can use the Solution section
instead of the Resulting Context to analyze this relation.
For example, let’s consider the pattern Navigation Tabs [9] (called P1)
which is Same as the pattern Navigation Tabs [7] (called P2). This relation is
given by the author of [9], so we consider it as correct and process the analy-
sis using our method. This latter starts by eliminating stop words [11] and
applying the Stemmer [12] on the elements of P1 and P2. After that, the
terms of these elements are weighted using the TFIDF method [13], and the
cosine similarity is calculated as explained in [10]. Also, our method checks
the inclusion [6], either it is true or false between each couple of elements.
We obtain the results shown in Table 1.
Table 1. Results of comparisons between patterns elements
Compared Elements Results
Similarity = 0,934
SC of P1 and SC of P2 SC of P1 includes SC of P2 = False
SC of P2 includes SC of P1 = False
Similarity = 0,956
RC of P1 and RC of P2 RC of P1 includes RC of P2 = True
RC of P2 includes RC of P1 = False
Similarity = 1
Forces of P1 and Forc-
Forces of P1 includes Forces of P2 = False
es of P2
Forces of P2 includes Forces of P1 = False
RC of P1 and SC of P2 Similarity = 0,136
RC of P2 and SC of P1 Similarity = 0,135
Since we obtain the similarity and inclusion results, we calculate the value of
each relation between P1 and P2, using the definition of each relation (Uses
[6] Refines [6], Same [10], Resulting-Starting [10], Starting-Starting [10]).
For instance, the relation Starting-Starting between the patterns P1 and P2 is
represented by the similarity value of the Staring Contexts of these patterns.
The results of the relations analysis are shown in Table 2.
Table 2. Results of relations analysis
Relationship Its value
P1 Uses P2 0
P1 Refines P2 0
P2 Uses P1 0
P2 Refines P1 0
Same 0,945
284
Starting-Starting 0,934
Resulting-Starting (P1 then P2) 0,136
Resulting-Starting (P2 then P1) 0,135
Finally, as in Kubo et al. method, the strongest relation of the eight types (P1
Uses P2, P1 Refines P2, P2 Uses P1, P2 Refines P1, Same, Resulting-
Starting (P1 then P2), Resulting-Starting (P2 then P1), Starting-Starting) is
assumed as the representative relationship. So we conclude in this case that
the patterns P1 and P2 are Same.
The relation Resulting-Starting : When the Solution of a pattern and the
Starting Context of another are similar, this means that the second pattern
(that we are interested in its Starting Context) can be applied after the first
pattern (that we are interested in its Solution), because the solution of the
first one provides the preconditions necessary to apply the second pattern.
So, we can use the Solution instead of the Resulting Context to analyze the
Resulting-Starting relation.
For example, let’s consider the patterns Titled Sections [14] (called P1)
and Closable Panels [14] (called P2) related by the Resulting-Starting rela-
tion. This relation is given by the author of these patterns, so we consider it
as correct and process the analysis using our method. We compare the ele-
ments of P1 and P2 and obtain the results shown in Table 3.
Table 3. Results of comparisons between patterns elements
Compared Elements Results
Similarity = 0,156
SC of P1 and SC of P2 SC of P1 includes SC of P2 = False
SC of P2 includes SC of P1 = True
Similarity = 0,119
RC of P1 and RC of P2 RC of P1 includes RC of P2 = True
RC of P2 includes RC of P1 = False
Similarity = 0,097
Forces of P1 and Forc-
Forces of P1 includes Forces of P2 = False
es of P2
Forces of P2 includes Forces of P1 = True
RC of P1 and SC of P2 Similarity = 0,248
RC of P2 and SC of P1 Similarity = 0,060
After that, we calculate the value of each relation between those patterns.
The results are shown in Table 4.
Table 4. Results of relations analysis
Relationship Its value
P1 Uses P2 0
P1 Refines P2 0
P2 Uses P1 0
P2 Refines P1 0,127
Same 0,137
285
Starting-Starting 0,156
Resulting-Starting (P1 then P2) 0,248
Resulting-Starting (P2 then P1) 0
Finally, considering the strongest relationship, we conclude that P1 and P2
are related via the relationship Resulting-Starting.
The relation Uses : When the Starting Context and the Solution of a pattern
are respectively included in the Starting Context and the Solution of another
pattern, then this means that the second pattern Uses the first one. So, we can
utilize the Solution instead of the Resulting Context to analyze this relation.
For example, let’s consider the pattern Extras On Demand [14] (called
P1) which Uses the pattern Closable Panels [14] (called P2) according to the
author of these patterns. So, we consider this relation as correct and process
the analysis via our method. We compare the different elements of P1 and P2
and obtain the results shown in Table 5.
Table 5. Results of comparisons between patterns elements
Compared Elements Results
Similarity = 0,216
SC of P1 and SC of P2 SC of P1 includes SC of P2 = True
SC of P2 includes SC of P1 = False
Similarity = 0,223
RC of P1 and RC of P2 RC of P1 includes RC of P2 = True
RC of P2 includes RC of P1 = False
Similarity = 0,130
Forces of P1 and
Forces of P1 includes Forces of P2 = True
Forces of P2
Forces of P2 includes Forces of P1 = False
RC of P1 and SC of P2 Similarity = 0,073
RC of P2 and SC of P1 Similarity = 0,114
After that, we calculate the value of each relation between those patterns.
The results are shown in Table 6.
Table 6. Results of relations analysis
Relationship Its value
P1 Uses P2 0,220
P1 Refines P2 0,173
P2 Uses P1 0
P2 Refines P1 0
Same 0,220
Starting-Starting 0,216
Resulting-Starting (P1 then P2) 0,073
Resulting-Starting (P2 then P1) 0,114
Finally, considering the strongest relationship, we conclude that the pattern
P1 Uses the pattern P2.
286
5 Conclusion and perspectives
Our way of looking at the analysis of relations between patterns is based on Kubo
et al method. We improved this method to enable it dealing with patterns which do
not give their Resulting Contexts in an explicit manner. Our idea consisted of using
the Solution section. As we have explained earlier, a such use does not alter the signif-
icance of the different relations treated. Some other improvements can be addressed to
face the drawbacks inherent to Kubo et al. method, and to offer more benefits for
patterns composition. Such as :
The block HTML Analysis of the method is limited to the treatment of pat-
terns expressed in HTML. This block can be extended to deal with patterns
expressed in other ways.
The method can be improved to treat patterns lacking of Starting Context
and/or Forces, which are two necessary elements to represent patterns in the
model of Kubo et al.
The method can be extended to offer the functionality of Patterns Retrieval,
which provides to a user having a particular problem, all available patterns
that treat this problem.
References
1. Zimmer, W.: Relationships between design patterns. In: PLoP (1994)
2. Conte, A., Fredj, M., Giraudin, J.P., Rieu, D.: P-Sigma : un formalisme pour une représentation
unifiée de patrons. In: Inforsid, Genève (2001)
3. Gnatz, M., Marschall, F., Popp, G., Rausch, A., Schwerin, W.: The Living Software
Development Process. Journal Software Quality Professional, Volume 5, Issue 3 (2003)
4. Henney, K.: Patterns Inside Out. Talk presented at Application Development, London (1999)
5. Crumlish, C., Malone, E.: Designing social interfaces. available at
http://www.designingsocialinterfaces.com/patterns/Main_Page (2009)
6. Hachemi, A., Ahmed-Nacer, M.: Primary inter-patterns relationships analysis. In: CARI2012,
Algiers (2012) unpublished.
7. Yahoo design pattern library. http://developer.yahoo.com/ypatterns/.
8. Develop effective XML documents using structural design patterns.
http://www.xmlpatterns.com/ (2000)
9. Lammi, J., Varjokallio, M., Hocksell, J.: A user interface design pattern library.
http://www.patternry.com.
10. Washizaki, H., Kubo, A., Takasu, A., Fukazawa, Y.: Analyzing Relations among Software
Patterns based on Document Similarity. In: IEEE Internationale Conference on Information
Technology : Coding and Computing (2005)
11. Salton, G., McGill, M.: Introduction to Modern Information Retrival. MG-Hill Inc., NY (1983)
12. Paice, C.: Another stemmer. SIGIR Forum, Vol. 24, No. 3, pp. 56–61 (1990)
13. Salton, G., Yang, C.: On the specification of term values in automatic indexing. Journal of
Documentation, Vol. 29, pp. 351–372 (1973)
14. Tidwell, J.: Designing interfaces - Patterns for effective interaction design. available at
http://designinginterfaces.com/firstedition/ (2005)
15. Prabhakar, T.V., Kumar, K.: Design Decision Topology Model for Pattern Relationship
Analysis. In: Asian PLOP 2010 Tokyo (2010)
287