=Paper=
{{Paper
|id=Vol-2191/paper29
|storemode=property
|title=Query Model and Similarity-Based Retrieval for Workflow Reuse in the Digital Humanities
|pdfUrl=https://ceur-ws.org/Vol-2191/paper29.pdf
|volume=Vol-2191
|authors=Lukas Malburg,Nicolas Münster,Christian Zeyen,Ralph Bergmann
|dblpUrl=https://dblp.org/rec/conf/lwa/MalburgMZB18
}}
==Query Model and Similarity-Based Retrieval for Workflow Reuse in the Digital Humanities==
https://ceur-ws.org/Vol-2191/paper29.pdf
Query Model and Similarity-Based Retrieval for
Workflow Reuse in the Digital Humanities
Lukas Malburg, Nicolas Münster, Christian Zeyen and Ralph Bergmann
Business Information Systems II
University of Trier
54286 Trier, Germany
[s4lumalb][s4nimuen][zeyen][bergmann]@uni-trier.de,
http://www.wi2.uni-trier.de
Abstract. Scientific Workflows do not seem to be broadly used today
in the Digital Humanities to perform text and data analysis. Although
they have become established in e-Science, modeling new workflows is
usually a demanding task, especially for novice users. Case-Based Rea-
soning (CBR) has been applied in the past to support the development of
workflows as an experience-based activity by retrieving past workflows. A
query language is needed for this purpose, but current languages do not
sufficiently consider different user groups and the information they can
provide. To address this issue, we present a query model to support novice
as well as experienced users. We identify common expression elements
from literature and integrate them in a prototypical CBR application
named Reuse Assistant to support workflow reuse in the RapidMiner
workflow tool. An experimental evaluation with non-expert users indi-
cates the potential of the Reuse Assistant to facilitate workflow reuse
and thus to simplify workflow development.
Keywords: Case-Based Reasoning · Workflow Reuse · Scientific Work-
flows · RapidMiner · Digital Humanities
1 Introduction
In recent years, a wide range of research tools have been emerged from vari-
ous text-oriented Digital Humanities (DH) projects, which can be seen in the
advent of tool and method collections such as TAPoR and Methodica1 . The ob-
vious goal of reusing such artifacts is not trivial to achieve since new research
questions usually require non-trivial and time-consuming adjustments or com-
binations of available tools [13]. Thus, modularization and reuse is a topic of
interest in DH for further research. In e-Science, Scientific Workflow Manage-
ment Systems (SciWFM) are widely used to create and execute workflows for
data analysis. The development of scientific workflows, however, can be a de-
manding and time-consuming task. This applies in particular for complex data
analysis that involve large amounts of data and require complex combinations
1
See http://tapor.ca and http://methodi.ca
2 L. Malburg et al.
of processing steps [3,15,16,17,21]. Consequently, sharing and reusing scientific
workflows is an important topic of research and various approaches have been
presented to support workflow reuse by searching available workflows [2,11,22].
Typically, these approaches require the user to specify the properties of a desired
workflow in a query. However, many query languages are composed of expression
elements that are difficult to understand for unexperienced users, thus restricting
their ability to find appropriate workflows [3].
Most recently, the benefits of using scientific workflows have also been recog-
nized in the context of DH [14]. Especially in text-oriented DH projects, scientists
can make use of numerous, well-established methods from the fields Natural Lan-
guage Processing (NLP) as well as Data and Text Mining. However, the usage
of scientific workflows in DH is still in its infancy. We assume that the more
a research domain is related to computer science, the easier it will be for the
respective scholar to get familiar with workflow modeling. Thus, a particular
challenge to support users in workflow reuse is posed by interdisciplinary DH
projects that, for instance, involve computer scientists and humanities scholars.
Among other aspects, we investigate this issue in the context of the eXplore! 2
project. In particular, we prove the practical application of the RapidMiner3
workflow tool for text analysis in the DH. A goal of this project is to accompany
the workflow creation and to develop a workflow modeling assistance to support
researchers in reusing past workflows.
This work is meant to be a first step towards the development of an assistance
by adapting and extending our past works on Process-Oriented Case-Based Rea-
soning (POCBR) to this new domain. In a nutshell, POCBR [1,18] integrates
CBR with process-oriented information systems and supports the development
of workflows as an experience-based activity. Experiential knowledge is repre-
sented in form of workflows and can be reused for similar problem situations.
POCBR does not require that a given user query matches exactly a workflow in
the case base. Instead, the most similar workflows are retrieved assuming they
can serve as a basis for creating a new one [19,23]. In this work, we address the
question of which expression elements should be available in a query language to
support users with different knowledge and experience in retrieving and reusing
scientific workflows. In a literature study, we identify expression elements that
are integrated in a new query model. The new query model is implemented in a
prototype named Reuse Assistant to support the reuse of RapidMiner workflows.
An experimental evaluation with users indicates the potential of the approach
to facilitate the reuse of workflows.
In the following, section 2 introduces different user groups working with Sci-
WFM as well as current assistance tools and their limitations. Section 3 presents
our approach for scientific workflow reuse by means of case-based reasoning and
section 4 describes the experimental evaluation while section 5 gives a conclusion
and discusses future work.
2
eXplore! is a cooperation project launched in 2016 with the Trier Center for Digital
Humanities (TCDH) at the University of Trier.
3
https://www.rapidminer.com/
Query Model and Retrieval for Workflow Reuse in the DH 3
2 Scientific Workflows in the Digital Humanities
User groups with varying experience in modeling workflows provide different in-
formation when they search for desired workflows. Cohen-Boulakia and Leser [3]
distinguish between “true users” and “power users”. The former are domain ex-
perts respectively domain scientists who constitute the largest group. Their aim
is primarily to analyze scientific data with SciWFM. They usually have no broad
experience in developing new workflows or analysis methods. Thus, true users
search workflows by specifying queries in a lower level of detail using keywords
or descriptions of the available input data or the desired output. We assume that
most DH scientists without an educational background in informatics rather be-
long to this group since they are not used to develop new workflows to perform
text and data analysis. In contrast, power users are workflow developers who
know how to create new workflows with SciWFM. Based on their broad ex-
perience they provide significantly more information and perhaps describe the
desired workflow, e.g. the topology in full or in part. Consequently, there should
be a continuum of query languages to express the properties of a desired work-
flow: On the one side, purely syntactical query languages enable to search for
simple keywords in a workflow description. These languages are mainly used
by true users in workflow repositories such as myExperiment4 . On the other
side, more expressive query languages enable users to specify the topology of a
workflow or to search for keywords in descriptions of tasks or within the entire
workflow. Such query languages are presumably more suitable for power users.
Cohen-Boulakia and Leser [3] emphasize that query languages targeting both
user groups are largely unexplored.
To the best of our knowledge, scientific workflows and SciWFM are not
broadly used in the DH. In contrast, in other domains such as e-Science, nu-
merous SciWFM such as KEPLER [16] are used to support analyses. Some of
the recent approaches in DH are tailored to certain application areas. For in-
stance, the CLARIN research infrastructure provides WebLicht5 , a web-based
tool for the composition of web services for tasks in NLP. WebLicht supports
users in creating analysis pipelines by only presenting the user web services that
are compatible in terms of input and output data constraints. However, only
very few example workflows are available and no reuse assistance is provided.
A few approaches exist for SciWFM that provide an assistance for work-
flow modeling based on available workflows. For instance, RapidMiner includes
the recommender system Wisdom of Crowds (WoC) [12] that suggests process-
ing steps and parameter settings. For this purpose, WoC stores and analyzes
the workflows created by users worldwide using machine learning techniques.
Based on these workflows, context-sensitive recommendations are continuously
displayed to users while they are modeling a workflow. WoC works fully auto-
matic and does not provide a query interface for users. Hence, it might be too
restrictive for more experienced users. Moreover, it suggests single processing
4
https://www.myexperiment.org/workflows
5
https://weblicht.sfs.uni-tuebingen.de/weblichtwiki/index.php
4 L. Malburg et al.
steps but no adaptations for the workflow under construction. Unfortunately,
the authors did not yet perform an evaluation that differentiates between novice
and expert users. The WINGS [6] SciWFM supports the generation of workflows
with an planning and semantic reasoning approach. Before WINGS generates
workflows, users create workflow templates by selecting components from cata-
logs in an editor and specify further properties or constraints using RDF triples.
It is also possible to reuse templates from other users. However, the large num-
ber of available workflow components for creating workflow templates and in
particular the use of RDF triples can be demanding, especially for true users.
Chinthaka et al. [2] present a generic CBR approach to assist users in reusing
scientific workflows. They use a keyword search and information about the work-
flow structure to retrieve appropriate workflows. However, the keyword search
just handles inputs as well as outputs from workflows and does not consider
other metadata.
3 Scientific Workflow Reuse by Case-Based Reasoning
We now describe our approach to support workflow reuse by means of POCBR.
This work extends our previous works on POCBR [1,19,23] by augmenting the
available query language with expression elements derived from literature.
3.1 Similarity-Based Retrieval of Workflows
A case in POCBR is usually a workflow that expresses experiential knowledge.
The approach is based on a workflow representation that uses semantically la-
operator type: naive_bayes
operator type: read_excel operator type: set_role operator type: filter_examples
name: Naive Bayes
name: Read Excel name: Set Role name: Filter Examples
parameter: laplace_correction = true
Filter
Read Excel Set Role Naive Bayes
Examples
DATA 1 DATA 2 DATA 3 DATA 4
data type: Data Table data type: Data Table data type: Data Table data type: Naive Bayes Model
role: intermediate role: intermediate role: intermediate role: global_output
input name: Read Excel output input name: Set Role output input name: Filter Examples output input name: Naive Bayes model
output name: Set Role input output name: Filter Examples input output name: Naive Bayes training set output name: Workflow output
task node data node controlflow edge dataflow edge semantic description
Fig. 1. NEST Graph Representation of a Data Mining Workflow
beled directed graphs named NEST graphs [1]. Figure 1 illustrates the NEST
graph of a data mining workflow created with RapidMiner for learning a Naive
Query Model and Retrieval for Workflow Reuse in the DH 5
Bayes classifier on a given Excel data file. A NEST graph consists of a set of
nodes (N) and edges (E) between nodes. Semantic descriptions (S) are domain-
dependent descriptions (key-value pairs) of nodes or edges. Additionally, each
node and edge has a specific type (T), e.g. task and data nodes as well as control-
flow and data-flow edges.
3.2 Expression Elements for Querying Scientific Workflows
To determine which expression elements are suitable to support the retrieval
and reuse of scientific workflows, we first carried out a literature study. Table 1
presents a selection of expression elements and papers from our literature study.
According to the categorization by Goderis et al. [9], expression elements are
grouped into “workflow structure” that define the topology of a workflow. El-
ements that relate to the entire workflow and its properties are grouped into
“workflow signature”. In the following, the term metadata refers to the workflow
signature. To support both user groups in retrieving and reusing scientific work-
Table 1. A Selection of Derived and Classified Expression Elements
G1 G2 G3 Others
Loops* [7,9,10] [5] [3,21] [22]
Conditionals* [7,9,10] [5] [3,21] [22]
Structure
Workflow
Data-flow* [7,9] [5] [3,21] [22]
Serviceflow / Control-flow* [7,9] [5] [3,21] [22]
Operators* [9] — [21] [22]
Generalized operators* — [4] [3] —
Restrictions of the topology — — [3] —
Input- and Output description [7,9,10] [4,5] [3,21] [22]
Workflow description* [8,10] — [3,21] [22]
Keywords / Tags* [8] — [3,21] [22]
Signature
Workflow
Reliability [7,8,9,10] — — [16]
Author [8,9] [5] [21] —
Rating by Community [8] [5] [21] —
Versioning [9,10] — [3] —
Title of the workflow* [8,10] — [21] —
flows, elements from the workflow structure are presumably more suitable for
power users while elements from the workflow signature are presumably more
suitable for true users. The expression elements marked with a star are imple-
mented in the Reuse Assistant to support the retrieval of RapidMiner workflows.
The papers are organized into groups (G) according to the researchers: G1 in-
cludes the authors of Goderis et al. [7,8,9,10], G2 the researchers of Gil et al.
[4,5] and G3 the authors of Starlinger, Cohen-Boulakia, and Leser [3,21]. Papers
that cannot be assigned to G1 , G2 , and G3 are combined in the group Others
[16,22]. This classification takes into account that research groups often discuss
the same expression elements.
6 L. Malburg et al.
3.3 Query Model and Case Workflow Representation
Our query model extends the Query Language for POCBR (POQL) by Müller
and Bergmann [19]. To support both user groups in retrieving and reusing sci-
entific workflows, we have integrated several expression elements from the lit-
erature study (see table 1). POQL is an expressive query language that al-
lows defining a desired workflow (DW) and one or more restriction workflows
(RW1 , ..., RWn ) with undesired elements. Therefore, it is possible to express
constraints on the graph structure [19]. Current research literature [3] states
that existing approaches for retrieving scientific workflows do not allow users
to express arbitrary constraints on the graph structure. Therefore, a restriction
part for a query language is desirable in order to express undesired workflow
properties in addition to the desired properties. The new query model differs
from POQL because in addition to the workflow structure metadata is impor-
tant, e.g. to support users who are maybe unable to express the structure of a
scientific workflow. Figure 2 illustrates the structure of a query based on POQL
with additional metadata.
Query Q = (Q+, Q)
Desired Workflow (DW) Desired Metadata (DM) Restriction Workflows (RW1, ..., RWn)
Workflow Structure Workflow Signature Workflow Structure Workflow Structure
(NEST Graph) (Metadata) (NEST Graph) (NEST Graph)
DW DMTags, DMTitle, DMKeys RW1 RWn
....
Q+ Q
Fig. 2. Query Structure for Similarity-Based Retrieval of Scientific Workflows
Based on previous work [19,23], a query Q = (Q+ , Q− ) contains a desired part
Q = (DW, DM ) and a restriction part Q− = (RW1 , ..., RWn ). DW represents
+
a desired workflow structure and Q− represents one or more undesired workflow
fragments. The desired metadata properties of a workflow are specified by DM ,
where DMT ags represents the searched tags of a query, DMT itle the desired
workflow title, and DMKeys the specified keywords.
The similarity assessment for the desired workflow structure and the restric-
tion workflow structures is similar to POQL in which a graph matching algorithm
by Bergmann and Gil [1] is used (see [19]). Furthermore, the literature study has
shown that POQL addresses all important structural properties for similarity-
based workflow retrieval. For this reason, it is not necessary to add further
structural elements and to consider them in the similarity assessment. However,
metadata should also be considered in the similarity assessment. Therefore, suit-
able expression elements should be part of the new query language.
With respect to the extended query model, the case workflow representa-
tion must be adapted accordingly. In addition to the workflow structure, the
Query Model and Retrieval for Workflow Reuse in the DH 7
NEST graph also contains metadata in the semantic description of task and data
nodes. For this reason, a strict separation between the workflow structure and
the workflow signature is not appropriate for case workflows. However, metadata
that characterizes the entire workflow should not be stored in the semantic de-
scription of specific nodes. Instead, this metadata should be handled separately
in the case workflow. The title of a workflow CWT itle , the tags of all operators
CWT ags , and the comments for documentation CWComment have to be taken
into account at workflow level. Although these metadata could be stored in the
semantic description of the workflow node (see [1]), we have decided to separate
the metadata from the workflow structure as much as possible since this fosters
the integration of other workflow representations that do not provide semantic
enrichments.
3.4 Similarity Assessment and Retrieval Process
The previously extended query model requires an adaptation of the similar-
ity computation presented in previous work (see [19,23]). With regard to the
added expression elements, the workflow signature, i.e., the metadata is the
new query part to be considered in the similarity assessment. We use a bag-of-
words approach to combine all comments in a workflow in a single set. This is
also applied for the tags of each operator. The similarity between the workflow
structures of the query (consisting of a desired workflow DW and one or more
restriction workflows RW1...n ) and a case workflow graph CWGraph is defined
by simP OQL (DW, RW1...n , CWGraph ) → [0, 1] (see [23] for more details). Please
note that arbitrary graph similarity measures can be used in this function. For
the whole query Q, the similarity to a case workflow CW is calculated as follows:
sim(Q, CW ) = simP OQL (DW, RW1...n , CWGraph ) ∗ w1
+ simtags (DMT ags , CWT ags ) ∗ w2
(1)
+ simtitle (DMT itle , CWT itle ) ∗ w3
+ simkeywords (DMKeys , CWComment ) ∗ w4
sim(Q, CW ) → [0, 1] is composed of several weighted and normalized similar-
ity measures. The weights are individually adjusted according to the application
scenario resulting in a sum of 1 (see section 4.1 for the weights chosen).
The tags DMT ags are arranged in a taxonomy. The similarity between a
query Q in which DMT ags represents the searched n tags (n = |DMT ags |) and
a case workflow CW with m tags (m = |CWT ags |) is specified as follows:
Xn
1
simtags (DMT ags , CWT ags ) = max {sim(DMT agi , CWT agx )} ∗ (2)
i=1
1≤x≤m n
sim(DMT agi , CWT agx ) → [0, 1] specifies the taxonomic similarity between T agi
from a query Q and T agx from the particular case workflow CW .
DMKeys refers to the keywords specified in a user query that are combined
into a bag-of-words. In order to enable an efficient retrieval, an index structure is
8 L. Malburg et al.
used similar to current web search engines. We use Apache Lucene6 to create the
index at the beginning of the retrieval process. Lucene tokenizes the comments of
the workflows and subsequently scans the index in main memory. By default, an
unnormalized similarity measure BM 25(DMKeys , CWComment ) → [0, ∞[ based
on Okapi BM25 (see [20] for more details) for a case workflow CW is returned
for a keyword search.
We define the similarity measure simkeywords (DMKeys , CWComment ) → [0, 1]
for a case base CB and a case workflow CW (CW ∈ CB) as follows:
BM 25(DMKeys ,CWComment )
max {BM 25(DMKeys ,CWComment )} hit
simkeywords (DMKeys , CWComment ) = ∀CW ∈CB (3)
0 else
Hit applies if Lucene has found one or more case workflows for the specified
keywords in the query.
For the similarity calculation of the workflow title, a similarity measure
simtitle (DMT itle , CWT itle ) → [0, 1] based on the Levenshtein distance is used.
4 Evaluation
In this section, we evaluate the Reuse Assistant according to two hypotheses:
H1 Using POCBR with the Reuse Assistant enables users to retrieve and
reuse suitable workflows for their current problem.
H2 Using POCBR with the Reuse Assistant supports users to solve a specified
problem faster than with the exclusive use of RapidMiner.
4.1 Evaluation Setup
For the evaluation, the Reuse Assistant is prototypically implemented based on
the CBR component provided by the CAKE framework7 . The graphical user
interface of the prototype is designed as follows: The specification of the title is
provided by an input field. To implement the tag search, all the tags available
in RapidMiner are extracted and made available to the user in a drop-down
menu. The workflow structure and comments created by the user are captured
within a XML representation generated by RapidMiner. The XML can be copied
and pasted from RapidMiner into the user interface. Once the query has been
specified, the user can search for a similar workflow. The similarity assessment
aggregates the weighted similarities of the individual expression elements. In a
prior experimental test phase, weights of 54 % for the graph structure and 46 %
for the metadata8 has been identified as suitable for the prototype. The most
similar workflow is then returned to the user as an image. Within the interface,
users can also view the next similar workflow or clear all entries.
6
https://lucene.apache.org/
7
See http://cake.wi2.uni-trier.de/
8
Composed of Title: 11 %, Tags: 25 %, and Keywords: 10 %
Query Model and Retrieval for Workflow Reuse in the DH 9
In addition to the operators included in RapidMiner, twelve generalized op-
erators are implemented as an extension for RapidMiner. These can be selected
by the user in the workflow modeling and thus can be used in the workflow struc-
ture of a query for retrieval. For this evaluation, the restriction workflow and the
parameter settings presented in section 3.3 have not been implemented within
the prototype. For simplicity purposes, the prototype has not been integrated
into the RapidMiner user interface.
To evaluate the hypotheses with the prototypical implementation, a case base
with 22 workflows is created with a total of 36 distinct operators and an average
of 5.8 operators per workflow. Furthermore, we semantically enrich the workflows
by adding a workflow documentation to the entire workflow and a descriptive
comment to each operator.
For the evaluation, we have invited participants from the “Data- and Web
Mining” lecture from the University of Trier for our evaluation. Eleven students
could be recruited that stated to have experience in creating RapidMiner work-
flows. The participants were randomly divided into two groups: six participants
in Group A and five in Group B. Group A worked exclusively with RapidMiner
and the Recommender WoC. In contrast, Group B worked with RapidMiner
and the developed Reuse Assistant. The evaluation was carried out separately
for each group. The Reuse Assistant or the Recommender WoC was introduced
to the respective group. During the familiarization phase, we also refreshed the
participant’s knowledge about workflow modeling with RapidMiner. Afterwards,
each student received four Data Mining tasks to solve. We have ensured that the
workflows retrieved from the case base do not completely solve the given prob-
lems and that adjustments are necessary in Group B. The computer screens were
recorded in order to determine in detail, which expression elements were used
and to measure the interaction times of each participant. Students were asked
to answer a survey after completing the tasks.
4.2 Results
All eleven participants were able to solve the respective tasks. The following
results obtained from the answers of Group B confirm hypothesis H1:
– In most cases (75 %), a suitable workflow could be found with the help of
the prototypical implementation (see statement a. in figure 3).
– All participants agreed that the prototypical implementation as an inte-
grated plugin in RapidMiner would be useful and that they would use it (see
statement b. in figure 3).
– As potential users of the prototypical implementation, the participants
stated that it is suitable for both true users and power users.
In the following, we present the results with respect to hypothesis H2. In
order to determine to what extent the participants could solve the tasks with
the prototype faster than with the exclusive use of RapidMiner, we measured
the time required to solve the problem. We can confirm that the participants
10 L. Malburg et al.
Fig. 3. Results of the Evaluation of Group B
of Group B (Avg. Time: 34:40 min. / SD: 7:02 min.) completed the four tasks
faster than those of Group A (Avg. Time: 36:07 min. / SD: 12:29 min.). In addi-
tion, it was observed that some participants of Group B had retrieved a suitable
workflow with the help of the prototype after a short time. At that moment,
however, the participants were not aware that they already had found the solu-
tion and were instead searching for further supposedly better workflows. If the
participants had analyzed the retrieved workflows more thoroughly, they would
have needed less time to solve the given problem. The fact that, unlike WoC, the
prototype is not integrated in the RapidMiner user interface and must therefore
be run as a second application presumably contributed to an extended period of
time until the problem was solved. The self-estimation of the participants from
Group B also confirms our hypothesis H2. Four out of five participants men-
tioned that they would save time in developing RapidMiner workflows, if they
would have the possibility to use the prototype as a plugin (see statement c. in
figure 3).
In addition, we have examined how often the provided expression elements
were used in the evaluation by the five participants of Group B. Since the expres-
sion elements could be combined, requests with several expression elements were
also possible. It was observed that the workflow structure was used in 96 % of
the 49 queries from Group B. Comments with keywords (47 %) and tags (67 %)
were also frequently used for searching. In contrast, the participants rarely used
the title of the workflow (6 %). However, this could be due to the fact that
the participants had no overview of the workflows in the case base and thus
the search for the title of a workflow turned out to be difficult. In practice, we
assume that the use of the title could be significantly higher.
5 Conclusion and Future Work
In this work, we identified expression elements from current research literature
and classified them according to the categorization by Goderis et al. [9]. Based on
Query Model and Retrieval for Workflow Reuse in the DH 11
this literature study, we have developed a first approach for case-based reuse of
scientific workflows named Reuse Assistant. The Reuse Assistant extends POQL
to include metadata, which is particularly useful for true users when searching
for scientific workflows. Furthermore, the similarity assessment is adapted to take
the metadata as well as the workflow structure into account. The evaluation indi-
cates that the Reuse Assistant enables true users as well as power users to reuse
scientific workflows and thus to facilitate the development process significantly.
Therefore, this work can be seen as a first step to support scientists from the
DH in using scientific workflows for data analysis. By regularly using the Reuse
Assistant, they can gain experience and become power users over time.
In the future, an extended evaluation particularly involving participants from
the DH using their own tasks is desirable to substantiate the indications found
in this work. Additionally, such an evaluation may discover further information
about the use of expression elements and the process of searching for scientific
workflows. An integration of the Reuse Assistant into RapidMiner as a plugin
is also desirable and could significantly decrease the time needed to develop
workflows. As explained previously, some participants have retrieved a suitable
workflow, but did not recognize it as a solution immediately. It would thus be
desirable to improve the presentation of the retrieval results. Whether the doc-
umentation of a case workflow should be enriched with further elements such as
parameter descriptions or help texts of operators could also be analyzed in future
work. Furthermore, the additional use of semantic technologies such as the use
of a thesaurus could make the keyword-based search more powerful. While this
project worked with fixed weights for the similarity measures, the importance of
the expression elements could be determined by the users themselves. Finally,
future research could investigate whether especially true users can be better sup-
ported in retrieving and reusing scientific workflows by a conversational POCBR
approach [23] in combination with our conceptual query model.
Acknowledgments. This work is funded by the German Federal Ministry of
Education and Research (BMBF, No. 01UG1606).
References
1. Bergmann, R., Gil, Y.: Similarity Assessment and Efficient Retrieval of Semantic
Workflows. Inf. Syst. 40, 115–127 (2014)
2. Chinthaka, E., Ekanayake, J., Leake, D.B., Plale, B.: CBR Based Workflow Com-
position Assistant. In: IEEE Congress on Services, Part I, SERVICES I 2009. pp.
352–355. IEEE Computer Society (2009)
3. Cohen-Boulakia, S., Leser, U.: Search, Adapt, and Reuse: The Future of Scientific
Workflows. SIGMOD Record 40(2), 6–16 (2011)
4. Gil, Y., González-Calero, P.A., Kim, J., Moody, J., Ratnakar, V.: A Semantic
Framework for Automatic Generation of Computational Workflows Using Dis-
tributed Data and Component Catalogues. J. Exp. Theor. Artif. Intell. (2011)
5. Gil, Y., Kim, J., Puga, G.F., Ratnakar, V., González-Calero, P.A.: Workflow
Matching Using Semantic Metadata. In: Proceedings of the 5th International Con-
ference on Knowledge Capture (K-CAP). pp. 121–128 (2009)
12 L. Malburg et al.
6. Gil, Y., Ratnakar, V., Kim, J., González-Calero, P.A., Groth, P.T., Moody, J.,
Deelman, E.: Wings: Intelligent Workflow-Based Design of Computational Exper-
iments. IEEE Intelligent Systems 26(1), 62–72 (2011)
7. Goderis, A.: Workflow Re-use and Discovery in Bioinformatics. Ph.D. thesis, School
of Computer Science, The University of Manchester (2008)
8. Goderis, A., Fisher, P., Gibson, A., Tanoh, F., Wolstencroft, K., Roure, D.D.,
Goble, C.A.: Benchmarking Workflow Discovery: A Case Study From Bioinformat-
ics. Concurrency and Comp.: Practice and Experience 21(16), 2052–2069 (2009)
9. Goderis, A., Li, P., Goble, C.A.: Workflow discovery: the problem, a case study
from e-Science and a graph-based solution. In: IEEE International Conference on
Web Services (ICWS). pp. 312–319 (2006)
10. Goderis, A., Sattler, U., Lord, P.W., Goble, C.A.: Seven Bottlenecks to Workflow
Reuse and Repurposing. In: 4th International Semantic Web Conference, Proceed-
ings. pp. 323–337 (2005)
11. Hauder, M., Gil, Y., Sethi, R.J., Liu, Y., Jo, H.: Making Data Analysis Expertise
Broadly Accessible through Workflows. In: WORKS’11, Proceedings of the 6th
Workshop on Workflows in Support of Large-Scale Science. pp. 77–86 (2011)
12. Jannach, D., Jugovac, M., Lerche, L.: Supporting the Design of Machine Learning
Workflows with a Recommendation System. TiiS 6(1), 8:1–8:35 (2016)
13. Kuhn, J., Reiter, N.: A Plea for a Method-Driven Agenda in the Digital Humani-
ties. In: Book of Abstracts of DH 2015 (2015)
14. Kuras, C., Eckar, T.: Prozessmodellierung mittels BPMN in Forschungsinfrastruk-
turen der Digital Humanities. In: Eibl, M., Gaedke, M. (eds.) INFORMATIK 2017.
pp. 1101–1112. Gesellschaft für Informatik, Bonn (2017)
15. Littauer, R., Ram, K., Ludäscher, B., Michener, W., Koskela, R.: Trends in Use of
Scientific Workflows: Insights from a Public Repository and Recommendations for
Best Practice. International Journal of Digital Curation 7(2), 92–100 (2012)
16. Ludäscher, B., Altintas, I., Berkley, C., Higgins, D., Jaeger, E., Jones, M.B., Lee,
E.A., Tao, J., Zhao, Y.: Scientific Workflow Management and the KEPLER System.
Concurrency and Computation: Practice and Experience 18(10), 1039–1065 (2006)
17. Ludäscher, B., Weske, M., McPhillips, T.M., Bowers, S.: Scientific Workflows: Busi-
ness as Usual? In: Dayal, U., Eder, J., Koehler, J., Reijers, H.A. (eds.) Business
Process Management, 7th International Conference, BPM, Proceedings. pp. 31–47.
LNCS, Springer (2009)
18. Minor, M., Montani, S., Recio-Garcı́a, J.A.: Process-oriented Case-based Reason-
ing. Inf. Syst. 40, 103–105 (2014)
19. Müller, G., Bergmann, R.: POQL: A New Query Language for Process-Oriented
Case-Based Reasoning. In: Bergmann, R., Görg, S., Müller, G. (eds.) Proceedings of
the LWA 2015. pp. 247–255. CEUR Workshop Proceedings, CEUR-WS.org (2015)
20. Robertson, S.E., Walker, S., Jones, S., Hancock-Beaulieu, M., Gatford, M.: Okapi
at TREC-3. In: Proceedings of The Third Text REtrieval Conference, Gaithers-
burg, Maryland, USA, November 2-4. pp. 109–126 (1994)
21. Starlinger, J.: Similarity Measures for Scientific Workflows. Ph.D. thesis, Humboldt
University of Berlin (2016)
22. Stoyanovich, J., Taskar, B., Davidson, S.: Exploring Repositories of Scientific Work-
flows. In: Proceedings of the 1st International Workshop on Workflow Approaches
to New Data-centric Science. pp. 7:1–7:10. ACM (2010)
23. Zeyen, C., Müller, G., Bergmann, R.: Conversational Process-Oriented Case-Based
Reasoning. In: Aha, D.W., Lieber, J. (eds.) Case-Based Reasoning Research and
Development - 25th International Conference, ICCBR 2017, Proceedings. pp. 403–
419. LNCS, Springer (2017)