Information Extraction from Legacy Spreadsheet-based

Information System

An Experience in the Automotive Context

Domenico Amalfitano

, Anna Rita Fasolino

, Porfirio Tramontana

, Vincenzo De Simone

Giancarlo Di Mare

and Stefano Scala

Dipartimento di Ingegneria Elettrica e Tecnologie dell’Informazione, Università Federico II, Via Claudio 21, Naples, Italy

Fiat Group Automobiles S.p.A.,Pomigliano Technical Center,Via ex Aeroporto, Pomigliano d’Arco, Italy

Keywords: Information Extraction, Legacy Systems, Spreadsheets, Automotive.

Abstract: Nevertheless spreadsheets were originally designed for computing purposes and for commercial

applications, they are often used in industry to implement Information Systems, thanks to the functionalities

offered by integrated scripting languages and ad-hoc frameworks (e.g., Visual Basic for Applications). This

technological solution allows the adoption of Rapid Application Development processes for the quickly

development of Spreadsheets-based Information Systems, but the resulting systems are quite difficult to be

maintained and very difficult to be migrated to other architectures such as Database-oriented Informative

Systems or Web applications. In this paper we present an approach for reverse engineering the data model

from an Excel spreadsheet-based information system. The approach exploits a set of heuristic rules that are

automatically applied in a seven-steps process. The applicability of the process has been shown in an

industrial context where it was used to obtain the UML class diagrams representing the conceptual data

models of three spreadsheet-based information systems.

1 INTRODUCTION

Spreadsheets are interactive software applications

designed for collecting and analyzing data in tabular

form. In a spreadsheet, data are organized in

worksheets, any of which is represented by a matrix

of cells each containing either data or formulas.

Modern spreadsheets applications (e.g. Microsoft

Excel) are integrated with scripting languages (e.g.,

Microsoft Visual Basic for Applications). These

languages allow the realization of interactive

functionalities for the management of the underlying

data by adopting Rapid Application Development

processes. In these scenarios, spreadsheets have left

behind their original role of calculation sheets,

becoming instead core business tools. As a recent

analysis showed, the usage of spreadsheets is

diffused in a relevant community made by many

millions of end-user programmers (Scaffidi, 2005).

Although it is possible to build Spreadsheets-

based Information Systems very quickly and without

high development costs, their quality is often low.

The data often present replication problems; data

management tends to be error prone due to the lack

of native consistency checking or data visibility

restriction mechanisms. Therefore, the user

dissatisfaction when working with these systems

may increase as the size and the complexity of data

increases, so business organizations may be

compelled to look for more effective solutions.

Migrating a legacy Spreadsheets-based

Information System towards new technologies and

platforms often provides an effective solution to

these problems. In the last years, several processes,

techniques and tools to support the migration of

legacy software systems have been proposed in the

literature (De Lucia, 2008; Bovenzi, 2003; Canfora,

2008).

The first and more critical step in a migration

process consists in the reverse engineering of the

data model of the information scattered in the cells

of different spreadsheets and worksheets. The

reverse engineered model will provide a useful

abstraction about the analyzed data and can be

considered the starting point for planning any

reengineering activity.

In (Amalfitano, 2014a) we presented a process for

migrating a legacy Spreadsheets-based Information

389

Amalﬁtano D., Fasolino A., Tramontana P., De Simone V., Di Mare G. and Scala S..

Information Extraction from Legacy Spreadsheet-based Information System - An Experience in the Automotive Context.

DOI: 10.5220/0005139603890398

In Proceedings of 3rd International Conference on Data Management Technologies and Applications (KomIS-2014), pages 389-398

ISBN: 978-989-758-035-2

 2014 SCITEPRESS (Science and Technology Publications, Lda.)

System to a new Web application. This process was

defined in an industrial context where we

reengineered a legacy system used in a tool-chain

adopted by an automotive company for the

development of embedded systems. According to

this company, such system was affected by many

maintenance and usability issues, which motivated

the migration process.

The first step of the migration process was

devoted to the reverse engineering of a data model

and of the business rules embedded in the

spreadsheet system. The successive steps regarded

the reengineering and reimplementation of the

system according to the target platform. The first

step was performed using a set of heuristic rules that

we described in (Amalfitano, 2014b).

In this paper we present a further refinement of

our work: we propose a structured reverse

engineering process that can be used to analyze

spreadsheet-based information systems and to infer a

UML class diagram from them. The class diagram

provides a conceptual model of the data embedded

in the spreadsheets and is comprehensive of classes,

class names, class attributes, association and

composition relationships. The process is based on

the execution of seven sequential steps that can be

automatically performed.

The effectiveness of the proposed process has

been assessed by a wider experimentation involving

more spreadsheets-based information systems from

the same industrial domain. In this paper we present

the proposed process and the results of the

experiment we performed.

The paper is organized as it follows: Section 2

presents related works, while Section 3 illustrates

the proposed data model reverse engineering

process. Section 4 shows an example of using the

process. Section 5 illustrates the experiment we

performed to evaluate the approach, while Section 6

presents some conclusive remarks.

2 RELATED WORKS

Due to the wide diffusion of spreadsheets in

business and in industry, a great interest in

spreadsheet analysis has been recently recorded.

Several authors addressed this problem and

proposed techniques and tools supporting the

automatic analyses of spreadsheets.

Several works in the literature aimed at inferring

a data model from spreadsheets. Some of them are

based on explicit extraction and transformation rules

that need to be provided by the users, such as the

technique proposed by Hung et al. (Hung, 2011).

The works of Abraham, Erwig et al. (Abraham,

2004; Abraham, 2006; Abraham, 2007; Abraham,

2009) exploit the existence of the shared template

underlying a given spreadsheet corpus, and propose

techniques for automatically inferring the templates.

The inferred template is hence exploited for safe

editing of the spreadsheets by end-users, avoiding

possible errors.

Two other approaches with similar purposes are

the ones of Mittermeir and Clermont (Mittermeir,

2002) and Ahmad et al. (Ahmad, 2003) that exploit

specific information about the type of data contained

in the spreadsheet cells.

Cunha et al. (Cunha, 2010), on the basis of the

previous results of Abraham and Erwig, propose

reverse engineering techniques to derive ClassSheet

models from existing spreadsheets by using data

mining techniques and infer functional dependencies

among columns.

Recently, Chen et al. (Chen, 2013) propose

automatic rules to infer some information useful in

the migration of a spreadsheet into a relational

database. They evaluated the effectiveness of the

proposed rules on a very large set including more

than 400 thousands of spreadsheets crawled from the

Web.

The work by Hermans et al. (Hermans, 2010)

addresses a problem very similar to ours that is the

recovery of a conceptual data model from

spreadsheets. That paper indeed presents a technique

for reverse engineering a data model from

spreadsheets that is based on two-dimensional

patterns. These patterns regard layout, data and

formulas included in the spreadsheets and can be

specified, recognized and transformed into class

diagrams. Some patterns were proposed by the

authors, as well as other ones were found in the

literature (e.g. Janvrin, 2000; Panko, 1994; Ronen,

1989). The technique has been validated with

respect to the spreadsheets corpus proposed by

Fisher and Rothermel (Fisher, 2005). Successively,

Hermans et al. (Hermans, 2011) evaluated the

usefulness of this approach by involving end users

from industry and abstracting leveled dataflow

diagrams, too.

However, these techniques are mainly applicable

to spreadsheets that are used as calculation sheets,

and that perfectly follow the predefined patterns. On

the contrary these techniques may not be effective

for analyzing spreadsheets not conforming to that

patterns or used as information systems. In the latter

case, indeed, no formula is present in the sheets and

DATA2014-3rdInternationalConferenceonDataManagementTechnologiesandApplications

390

the rules used by the authors to classify cells may be

misleading.

The spreadsheets developed in the considered

industrial context were used as information systems

too and were not calculation sheets. As a

consequence, we were not able to reuse the

techniques found in the literature as-are, but we had

to adapt them to the specific context. In particular,

we had to look for heuristic rules applicable to the

considered types of spreadsheets and we had to

define a process made of a sequence of steps for

applying the selected rules. The process we defined

will be illustrated in the following section.

3 THE CONCEPTUAL DATA

MODEL REVERSE

ENGINEERING PROCESS

The industrial context of our work included a large

number of spreadsheets, implemented by Excel files,

used in the development process of Electronic

Control Units (ECU) in the automotive company.

They supported the Verification & Validation

activities (V&V) and the management of Key

Performance Indicators (KPI) about the development

process. The spreadsheets inherited from a same

template and included some VBA functionalities

providing data entry assistance. Moreover, their

cells followed well-defined formatting rules (i.e.,

font, colors, and size of cells) that improved the

readability and usability of the spreadsheets, and

complied to the following layout rule: all the data

concerning the same topic in a single spreadsheet

file were grouped together in rows or columns

separated by empty cells or according to specific

spreadsheets patterns (Hermans, 2010).

Taking the cue from other works proposed in the

literature (Cunha, 2010; Hermans, 2010; Abraham,

2006), we founded our data model reverse

engineering process on a set of heuristic rules. We

defined a process made of seven steps that can be

automatically performed in order to infer, with

gradual refinements, the UML class diagram of the

considered information system. In each step, one or

more heuristic rules are executed. Each rule is based

on the analysis of one or more spreadsheets

belonging to the corpus of spreadsheet files

composing the subject information system. In the

following we describe the steps of the process.

Step 1: in this step we preliminarily apply Rule 1

that abstracts a class named Sp whose instances are

the Excel files that comply to a same template.

Step 2: in this step we exploit the Rule 2 that is

executed on a single spreadsheet file of the corpus.

The rule associates each non empty sheet of the

Excel file with a class S

, having the same name of

the corresponding sheet. Moreover, an UML

composition relationship between the Sp class and

each S

belonging to the file is inferred. The

multiplicity of the association on each S

side is

equal to 1. Figure 1 shows an example of applying

this rule on an example spreadsheet.

Figure 1: Example of Step 2 execution

Step 3: in this step we exploit the Rule 3 that is

executed on a single spreadsheet file of the corpus.

This heuristic, according to (Abraham, 2006;

Hermans, 2010), associates a class A

for each non

empty cell area Area

of a sheet already associated to

a class S

by the Rule 2. Each class A

is named as

follows: Name of the sheet_Areaj. Moreover an

UML composition relationship between the S

class

and each A

class is inferred. The multiplicity of the

association on each A

side is equal to 1. Figure 2

shows an example of Step 3 execution.

Step 4: In this Step two rules Rule 4.1 and Rule

4.2 are sequentially executed.

The Rule 4.1 is applied to discriminate the

header cells (Abraham, 2004) of each area Area

that

was inferred in the previous step. Rule 4.1 analyzes

the content of the spreadsheet files belonging to the

corpus in order to find the invariant cells for each

area Area

. An invariant cell of an area is a cell

whose formatting and content is the same in all the

analyzed spreadsheets.

The set of invariant cells of an area Area

composes the header of that area. Figure 3 shows

how the Rule 4.1 works. In a first phase (Figure 3-A)

all the spreadsheets are queried to select, for each of

them, the cells of a given area. In the next phase

(Figure 3-B) all the selected cells are analyzed to

recognize the invariant cells for the considered area.

Intuitively, if we imagine to overlap the contents of

InformationExtractionfromLegacySpreadsheet-basedInformationSystem-AnExperienceintheAutomotiveContext

391

Figure 2: Analysis of non-empty cell areas belonging to

Sheet1 executed in Step 3.

of all the spreadsheets for a given area Area

, then

the header is given by the cells that, for the

considered area, are invariant in all the files, as

shown in Figure 3-B.

Rule 4.2 is executed on the headers inferred by

Rule 4.1. For each area Area

, this heuristic analyzes

the style formatting properties of the cells

composing its header. It permits to discriminate

subareas SubArea

having header cells satisfying

specific patterns. Rule 4.2 associates a class SA

for

each SubArea

. Each class SA

is named as follows:

Name of the sheet_SubAream, if no name could be

associated to the class according to the recognized

patterns. The names of the attributes of the class are

inferred from the values contained into the header

cells.

Moreover, an UML composition relationship

between the class S

and each related SA

class is

inferred. The multiplicity of the association on each

class SA

side is equal to 1. Some examples of Rule

4.2 executions are reported in Figure 4, 5, 6 and 7.

As an example in Figure 4 a UML class, having

the default name Sheet1_Area1, is inferred since

three consecutive header cells have the same

formatting style. The attributes of the class are

named after the values contained into the header

cells.

Figure 5 shows a variant of the example reported

in Figure 4. In this case two UML classes are

inferred since two groups of consecutive cells

having the same formatting characteristics were

found.

A further pattern is shown in Figure 6 where the

header is structured in two different levels. The

upper level, composed of merged cells, permits to

infer a class named ClassA, while its attributes are

named after the values contained in the header cells

of the lower level.

In the example shown in Figure 7, the previous

pattern is applied twice and a composition

relationship is inferred between the two obtained

classes.

Step 5: in this step we exploit Rule 5 that is

applied on the whole corpus of spreadsheets.

Figure 1: Execution of Step 3.

Figure 4: Example of header cells pattern inferring a

single class and its attributes.

Rule 5 is applied to all the subareas SubArea

find possible sub-subareas made by group of data

cells having the same style formatting properties. If

a subarea SubArea

is composed by two or more

sub-subareas then the heuristic associates a class,

SSA

for each sub-subarea. Each class SSA

is named

as follows: Name of the sheet_SubSubAreak.

Moreover, the new classes substitute the class

associated to the SubArea

, and an UML

composition relationship between the A

class and

each SSA

class belonging to the area is inferred. The

multiplicity of the association on each class SSA

side is equal to 1. Figure 8 shows an example of how

Step 5 works.

Step 6: in this step the Rule 6 is exploited. This

heuristic is applied to the overall corpus of

spreadsheets. It analyzes the formatting properties of

the cells belonging to consecutive subareas and sub-

subareas in order to infer association relationships

and multiplicities between the classes that were

associated to these areas in the previous steps.

DATA2014-3rdInternationalConferenceonDataManagementTechnologiesandApplications

392

An example of how this rule works is reported in

Figure 9. In this case two consecutive subareas are

considered, i.e., SubArea1 related to columns A and

B and SubArea2 corresponding to columns C and D.

The classes named Sheet1_SubArea1 and

Sheet1_SubArea2 were associated to SubArea1 and

SubArea2, respectively. Since for each spreadsheet a

tuple of SubArea1 is composed by a number of

merged cells that is an integer multiple of the

number of merged cells related to a tuple of

SubArea2, then Rule 6 infers a UML association

between the two classes related to the two subareas.

The multiplicity of the association on the class

side related to the class Sheet1_SubArea1 is equal to

1 whereas the one on the other side is 1..*.

Step 7: in this step the Rule 7 is exploited. This

heuristic is applied to the overall corpus of

spreadsheets. It analyzes the value of the cells in

order to infer association relationships between

classes. As an example, if in all the spreadsheets, for

each cell of a column/row that was exploited to infer

an attribute of a class A there is at least a cell of a

column/row that was exploited to infer an attribute

of a class B having the same value, then it is

possible to define a UML relationship between the

UML Class A and the UML Class B.

Moreover, if the cells of the column A have

unique values then the A side and B side

multiplicities of the inferred UML relationship are 1

and 1..* respectively, as shown in Figure 10.

Figure 5: Example of header cells pattern inferring two

classes and their attributes.

Figure 6: Example of header cells pattern inferring a

single class with attributes and class name.

Figure 7: Example of header cells pattern inferring two

classes with attributes, class names and their composing

relationship.

Figure 8: Example of Step 5 Execution.

Figure 9: Example of Step 6 Execution.

Figure 10: Example of Step 7 Execution.

InformationExtractionfromLegacySpreadsheet-basedInformationSystem-AnExperienceintheAutomotiveContext

393

4 CASE STUDIES

To analyze the effectiveness of the proposed

process, we performed a series of case studies

involving more spreadsheets-based information

systems from the same industrial domain. The goal

of these case studies was to evaluate the

applicability of the process and the acceptability and

precision of the inferred models.

In the first case study we used the proposed

process to reverse engineer the conceptual data

model from a legacy Spreadsheets-based

Information System implemented in Microsoft

Excel. This system was used by the HIL

(Hardware-in-the-Loop) Validation Team of Fiat

Group Automobiles and provided strategic support

for the definition of high-level testing specifications,

named Test Patterns. Test Patterns represent

essential artifacts in the overall testing process

(Shokry, 2009) since they allow the automatic

generation of test cases, the Test Objects, necessary

to exercise the Electronic Control Units (ECUs) of

automotive vehicles. The generation process is

carried out thanks to a lookup table that embeds both

the translation of high-level operations to the

corresponding set of low-level instructions, and the

mapping between input and output data to the set of

ECU pins.

Test Patterns were implemented by means of a

predefined Excel template file organized into 7

worksheets, referred to different phases of the

testing process.

The worksheets embedded the model of the data.

In the considered context, such template has been

adopted to instantiate a set of 30,615 different Excel

files, constituting the overall spreadsheets-based

informative system. These spreadsheets contained

on average 2,700 data cells.

All of these files inevitably shared the same

structure (i.e., the data model) with a high rate of

replicated data (about 20% of data cells recurred

more than 1,300 times, while about the 50% more

than 100 times). This high rate of replication was

mainly due to the fact that data were scattered

among multiple spreadsheets. Therefore the

underlying model was not normalized, with any

information about related data. In addition, the

automatic verification of data consistency and

correctness was not natively supported.

At the end of the process execution we obtained

a UML class diagram composed of 36 classes, 35

composition relationships and 23 association

relationships. Seven of these classes are associated

with the worksheets composing each spreadsheet

file, while the remaining ones are related to the areas

and subareas of each worksheet. Figure 11 shows the

conceptual data model class diagram that was

automatically inferred by executing the process. For

readability reasons, we have not reported all the

association relationships and the associations’

multiplicities, whereas we reported all the classes

and the composition relationships that were inferred.

Figure 12 shows an example of rules execution

on the Test Patterns sheet that is the most complex

one. By executing the rules we were able to infer

five classes, 4 association relationships and 5

composition relationships. Moreover, it shows how

the Rule 4 were able to infer the two classes Test

Step and Expected Results and Rule 6 proposed the

association relationship between them.

In the second and third case study, we analyzed

two further systems used by the company. The

systems included a Software Factory Test Case

(called SW TC) repository and a KPI repository

(hereafter KPI).

The SW TC repository contains essential

artifacts of the overall testing process in the

considered company, since they allow the automatic

generation of executable MATLAB test scripts

necessary to validate the ECU models. This

information system is composed by 14,000 Excel

files inheriting from a common template composed

by 10 sheets. These spreadsheets contained 4,000

data cells on average. About the 75% of data cells

are replicated in the spreadsheets.

KPI repository contains the key performance

indicators of each development project regarding a

software component belonging to a specific type of

vehicle. The information system is composed by

1,500 Excel files inheriting from a common template

composed by 8 sheets. These spreadsheets contained

1,370 data cells on average. About the 77% of data

cells are replicated in the spreadsheets.

Table 1 shows the results of the conceptual data

model reverse engineering process involving the

three case studies. It shows the number of classes

and relationships that were automatically inferred

for each information system.

After the analysis, we performed a validation

step in order to assess the acceptability and the

precision of the models inferred by the reverse

engineering process. To this aim we enrolled three

experts from the company belonging to the

application domains of the information systems.

We submitted them: (1) the inferred data models,

(2) a report containing the description of each

inferred UML item and one or more traceability

DATA2014-3rdInternationalConferenceonDataManagementTechnologiesandApplications

394

Table 1: Reverse Engineering Results.

Information

System

#Classes

#Association

Relationships

#Composition

Relationships

SW TC 49 33 48

Test

Pattern

36 23 35

KPI 25 12 24

links towards the spreadsheet’s parts from which it

was extracted, (3) a set of questions in order to

collect the expert judgments about the validity of the

UML item proposals. We asked the experts to

answer the questions and thus we were able to assess

the effectiveness of the overall reverse engineering

process by means of the Precision metric reported

below:

 

. . .

. .

 100

I.E. is the set of the UML elements, i.e., classes,

class names, class attributes, relationships between

classes and multiplicities of the relationships, that

were inferred by the process.

. .

is the cardinality

of the I.E. set.

. . .  ⊆ .. is the number of the inferred

UML elements that were validated by the industrial

expert and

. . .

is the cardinality of this set.

Table 2 shows the precision values we obtained for

each information system.

Table 2: Evaluation Results.

Information System Precision

SW TC 81%

Test Pattern 84%

KPI 92%

The precision values reported showed that more

than 80% of the inferred elements were validated by

the experts.

In the remaining cases, the experts proposed

some minor changes to the candidate classes. As an

example, with respect to the first system, the Test

Patterns one, the expert decided to candidate

additional classes by extracting their attributes from

the ones of another candidate class, or to merge

some of the proposed classes into a single one.

In particular, with respect to the class diagram

reported in Figure 12, he proposed to (1) discard the

PlotOutput class and to move its attributes into the

ExpectedResult one and (2) to extract a new class,

named Repetition, having the attributes repetition

and eventTime that were given from the candidate

class named Test Step. Definitely, the expert

proposed 1 class split and 2 classes merges

operations. Similar changes were proposed also in

the other two case studies we performed.

Moreover, we further analyzed the results of the

validation steps, in order to assess the applicability

of the rules used throughout the process. In this way,

we were able to learn some lessons about the

heuristic rules.

Figure 11: Inferred conceptual UML class diagram for the Test Pattern Information System.

InformationExtractionfromLegacySpreadsheet-basedInformationSystem-AnExperienceintheAutomotiveContext

395

Figure 12: Example of class proposal and its mapping with the spreadsheet file.

As to the rules regarding the identification of

classes, we observed that they were always

successful except for three cases. In two cases, the

expert decided to merge two candidate classes into a

single one. These classes had been considered as

different by Rule 4.2, since they derived from two

subareas having headers with different colors but

actually belonging to the same concept.

In a single case the expert decided to extract an

additional class from a candidate one and assigned it

a subset of the attributes of the inferred one. In this

case the Rules 4.2, 5 and 6 were not able to identify

this extra class since the cells associated to these two

concepts had the same formatting and layout style.

As a consequence, our process was not able to

discriminate between them. In both cases the process

failed because the spreadsheet did not comply with

the formatting rules exploited by our process.

As to the rules we used to associate the candidate

classes with a name, only in 21 over 110 cases they

failed, since the spreadsheets did not include

meaningful information to be exploited for this aim.

Furthermore we observed that in some cases the

expert decided to discard some of the proposed

composition relationships between the inferred

classes. This fact occurred when the proposed

conceptual class diagram presented a particular

pattern like the one showed in Figure 13. In this

case, the expert decided to move the attributes of the

leaf class (Sheet_SubArea1) to the Sheet1 class and

to remove the remaining classes, as shown in Figure

14. This specific pattern occurred 8 times and in 6

cases the expert decided to make these changes. This

result showed us the need to introduce new rules

aimed at reducing the complexity of the class

diagram that may be used in the occurrence of this

particular pattern.

Figure 13: Conceptual class diagram inferred from a sheet

having a single area containing only one subarea.

DATA2014-3rdInternationalConferenceonDataManagementTechnologiesandApplications

396

Figure 14: Class Diagram reduction proposed by the

domain expert.

5 CONCLUSIONS

In this paper we presented a process for inferring a

conceptual data model from a spreadsheet-based

information system. The process has been defined in

an industrial context and validated by an experiment

involving three different spreadsheets-based

information systems from the considered automobile

industrial domain. The results of the experiment

showed the applicability of the process and the

acceptability of the inferred models, according to the

judgment of experts about the application domain of

the spreadsheets.

Our work differ from other ones described in the

literature, since the proposed approach has been

tailored to spreadsheets used to implement

information systems, rather than calculation sheets.

In future work, we plan to perform further

experimentations involving other spreadsheets-based

systems belonging to different application domains.

Moreover, we want to extend our approach further,

by proposing reverse engineering techniques aimed

at inferring the functional dependencies between the

data embedded in the spreadsheets by analyzing the

VBA functionalities they include.

ACKNOWLEDGEMENTS

This work was carried out in the context of the

research project IESWECAN (Informatics for

Embedded SoftWare Engineering of Construction

and Agricultural machiNes - PON01-01516),

partially founded by the Italian Ministry for

University and Research (MIUR).

REFERENCES

Abraham R. and Erwig M., Header and unit inference for

spreadsheets through spatial analyses. In Proceedings

of the IEEE International Symposium on Visual

Languages and Human-Centric Computing

(VL/HCC), 2004, pages 165–172.

Abraham R. and Erwig M., Inferring templates from

spreadsheets. In Proceedings of the 28th International

Conference on Software Engineering (ICSE), ACM,

New York, NY, USA, 2006, pages 182–191.

Abraham R., Erwig M. and Andrew S., A type system

based on end-user vocabulary. In Proceedings of the

IEEE Symposium on Visual Languages and Human-

Centric Computing (VL/HCC), Washington, DC,

USA, IEEE Computer Society, 2007, pages 215–222.

Abraham R. and Erwig M., Mutation operators for

spreadsheets. IEEE Transactions on Software

Engineering, 35(1):94–108, 2009.

Ahmad Y., Antoniu T., Goldwater S. and Krishnamurthi

S., A type system for statically detecting spreadsheet

errors. In Proceedings of the IEEE International

Conference on Automated Software Engineering,

2003, pages 174–183.

Amalfitano D., Fasolino A.R., Maggio V., Tramontana P.,

Di Mare G., Ferrara F., Scala S., Migrating legacy

spreadsheets-based systems to Web MVC architecture:

An industrial case study, Proceedings of CSMR-

WCRE, 2014, pages 387-390.

Amalfitano D., Fasolino A.R., Maggio V., Tramontana P.,

De Simone V., Reverse Engineering of Data Models

from Legacy Spreadsheets-Based Systems: An

Industrial Case Study, Proceedings of the 22nd Italian

Symposium on Advanced Database System, 2014,

pages 123-130.

Bovenzi D., Canfora G., Fasolino A.R., Enabling legacy

system accessibility by Web heterogeneous clients. In

proceedings of the Seventh European Conference on

Software Maintenance and Reengineering, IEEE CS

Press, 2003, pages 73-81.

Canfora G., Fasolino A.R., Frattolillo G., Tramontana P.,

A wrapping approach for migrating legacy system

interactive functionalities to Service Oriented

Architectures. Elsevier, Journal of Systems and

Software, 2008, vol. 81(4):463–480,

Chen Z. and Cafarella M., Automatic web spreadsheet

data extraction. In Proceedings of the 3rd International

Workshop on Semantic Search Over the Web (SS@

'13). ACM, New York, NY, USA, 2013, 8 pages.

Cunha, J., Erwig M., Saraiva J., Automatically Inferring

ClassSheet Models from Spreadsheets. In IEEE

Symposium on Visual Languages and Human-Centric

Computing (VL/HCC), IEEE CS Press, 2010, pages

93-100.

De Lucia A., Francese R., Scanniello G., Tortora G.,

Developing legacy system migration methods and

tools for technology transfer. In Software Practice and

Experience 38(13), Wiley,2008, pages 1333-1364.

Di Lucca G.A., Fasolino A.R., De Carlini U., Recovering

class diagrams from data-intensive legacy systems. In

InformationExtractionfromLegacySpreadsheet-basedInformationSystem-AnExperienceintheAutomotiveContext

397

Proceedings of International Conference on Software

Maintenance, ICSM, IEEE CS Press, 2000, pages 52-

62.

Fisher M. and Rothermel G., The EUSES spreadsheet

corpus: A shared resource for supporting

experimentation with spreadsheet dependability

mechanisms. In In 1st Workshop on End-User

Software Engineering, 2005, pages 47–51.

Hermans F., Pinzger M., van Deursen A., Automatically

extracting class diagrams from spreadsheets. In

proceedings of the 24th European conference on

Object-oriented programming (ECOOP'10). Springer-

Verlag, Berlin, Heidelberg, 2010, pages 52-75.

Hermans F., Pinzger M. and van Deursen A., Supporting

professional spreadsheet users by generating leveled

dataflow diagrams. In Proceedings of the 33rd

International Conference on Software

Engineering (ICSE '11). ACM, New York, NY, USA,

2011, pages 451-460.

Hung V., Benatallah B. and Saint-Paul R., Spreadsheet-

based complex data transformation. In Proceedings of

the 20th ACM international conference on Information

and knowledge management (CIKM '11). ACM, New

York, NY, USA, 2011, pages 1749-1754.

Janvrin D. and Morrison J., Using a structured design

approach to reduce risks in end user spreadsheet

development. Information & Management, 37(1):1–

12, 2000.

Mittermeir R. and Clermont M., Finding high-level

structures in spreadsheet programs. In Proceedings of

the Ninth Working Conference on Reverse

Engineering (WCRE), IEEE Computer Society,2002,

pages 221-232.

Panko R.R. and Halverson R.P., Individual and group

spreadsheet design: Patterns of errors. In Proceedings

of the Hawaii International Conference on System

Sciences (HICSS), 1994, pages 4–10.

Ronen B., Palley M.A. and Lucas H.C., Spreadsheet

analysis and design. Communications of the ACM,

32:84–93, 1989.

Scaffidi C., Shaw M., Myers B., Estimating the Numbers

of End Users and End User Programmers. In

Proceedings of the 2005 IEEE Symposium on Visual

Languages and Human-Centric Computing, 2005,

pages 207-214.

Shokry H., Hinchey M., Model-Based Verification of

Embedded Software. In IEEE Computer, 42(4), 2009,

pages 53-59.

DATA2014-3rdInternationalConferenceonDataManagementTechnologiesandApplications

398