Business Process Search within Unstructured Repositories
Maya Lincoln and Avi Wasser
ProcessWeb, Haifa, Israel
Keywords: Business Process Search, Process Ontology, Business Process Repositories, Natural Language Processing,
Semantic Search, Operational Search.
Abstract: In recent years, researchers have become increasingly interested in developing frameworks and tools for
searching business process model repositories. While research on searching structured repositories has been
extensive, little attention was dedicated to searching business process content within unstructured
repositories, such as the Web. We demonstrate why current search technologies are not useful for extracting
process content from the Web, and explain the core reasons for the deficiency. We then present a framework
for overcoming this material weakness, and discuss possible applications for realizing the suggested
method.
1 INTRODUCTION
Business Process Models (BPMs) are considered an
important mine of organizational knowledge, and
therefore are a major source for searching and
retrieving operational and enterprise related data.
Researchers have become increasingly interested
in developing methods and tools for retrieving
information from business process repositories
(Awad, 2008; Lincoln and Gal, 2011; Wasser et al.,
2006). While research on searching structured
repositories has been extensive, little or no attention
was dedicated to searching business process content
within unstructured repositories, such as the Web.
Such repositories are constantly becoming more
extensive, and are accessible to a wide user
population through search engines.
Two common methods for retrieving information
from a repository are querying and searching.
Querying is aimed at retrieving information using a
structured query language. The significance of
querying business processes has been acknowledged
by BPMI that launched a Business Process Query
Language (BPQL) initiative. Searching, on the other
hand, allows information retrieval using keywords or
natural language and was shown to be an effective
method for non-experts.
Research in the field of business process retrieval
has mainly focused on semantics and structural
similarity analysis techniques (Awad, 2008;
Markovic et al., 2008; Beeri et al., 2008; Karni et al.,
2014). Using these frameworks one can retrieve
process models that either contain semantically
related components (e.g. activity names with a
specified keyword) or match a requested graph
structure: e.g. that presents a sequence of activities.
While these methods can be applied on structured
process repositories, it is practically impossible to
apply them on the unstructured Web.
Figure 1: An example of search results for “how to claim a
tax refund.”
In order to illustrate why semantic search is not
adequate for process retrieval from unstructured
repositories, we will present a motivating example,
as follows. Consider an employee interested in
finding out “how to claim a tax refund.” An
expected outcome of this retrieval request would be
a process model that represents the order of
activities that one should follow in order to achieve
the required process goal, as illustrated in Fig. 1a.
The benefit of such a retrieval framework is that the
result is ready for execution. Without any
preliminary knowledge of the underlying repository
structure, the user can receive a full-fledged process
467
Lincoln M. and Wasser A..
Business Process Search within Unstructured Repositories.
DOI: 10.5220/0005169404670474
In Proceedings of the International Conference on Knowledge Engineering and Ontology Development (KEOD-2014), pages 467-474
ISBN: 978-989-758-049-9
Copyright
c
2014 SCITEPRESS (Science and Technology Publications, Lda.)
model.
The retrieval output is related to the search
phrase in operational terms. For example, Fig. 1a
provides a segment that is not similar semantically
to the search phrase text. Specifically, all three
search phrase terms (“Claim”, “Tax” and “Refund”)
are not represented by any of its activities. Such
“how-to” questions are hard to fulfill using common
query languages due to the complex logic that is
embedded within such questions (Beeri et al., 2008)
and especially without specific knowledge on
process structure and activity naming. Therefore,
using querying techniques on the Web, would yield
a list of data items (e.g. Web pages, or media items)
with semantically similar titles, as illustrated in Fig.
1b. Such outcome does not tell the user “how-to”
fulfill the process goal in a structured and
operational manner.
In this work we present a framework that aims to
overcome this material weakness. The suggested
retrieval framework is based on operational, and not
on semantic similarities. This is done by processing
the unstructured Web content, into a structured
process repository. Then, the operational business
logic is extracted from the generated repository
through the analysis of process components. The
proposed framework dynamically extracts process
models according to the ad-hoc requests as
expressed in the user's search phrase.
This work proposes an innovative method for
searching business process models within the Web,
while making use of the how-to knowledge that is
encoded in this valuable repository. The following
contributions are presented: (a) automatic
construction of processes from unstructured, non-
BPM, repositories; (b) generic support to an
operation-based search of business process models
within the Web; and (c) capability to generate ad-
hoc process model results from natural language or
graph-based Web search.
The rest of the paper is organized as follows: we
present related work in Section 2, positioning our
work with respect to previous research. In Section 3
we present the major shortcomings of current Web
search engines in extracting process data. Section 4
formulates the search problem and describes our
framework for searching business processes within
the Web. We discuss future research elaborations
and conclude in Section 5.
2 RELATED WORK
Related works include query and search techniques
in BPM. Works such as Shao et al. (2009) and Awad
(2007) query business process repositories to extract
process model (graph) segments. Such methods
require prior knowledge of the structure of the
process repository and the exact notation that is used
to express it. Therefore, they are not adequate for
search on the Web that should work well even
without prior knowledge regarding the process
repository.
Keyword search on general tree or graph data
structures can also be applied to process repositories
(Hristidis et al., 2003; Guo et al., 2003; He et al.,
2007). Some works extend the tree and graph
keyword search methods to support a more intuitive
interface for the user by enabling searches based on
natural language (Katz et al., 2010). The retrieved
information in both keyword and natural language
search methods is in the form of single process
model components such as activities and roles that
are semantically similar to the searched phrase.
These techniques are merely relevant to process
search on the Web, since in this case (a) users are
seeking to receive a complete process; and (b) the
expected process result is usually not related
semantically to the search phrase, but rather
operationally. The work of Lincoln and Gal (2011)
extends the above line of works. This work supports
the retrieval of complete process segments by
applying dynamic segmentation of the process
repository. The search result is a compendium of
data (a segment of a business process model) related
to the operational meaning of the searched text.
Nevertheless, as all other works, this method relies
also on a process-structured database, and cannot
work “as is” on an unstructured repository, such as
the Web.
Another line of work focuses on automatic
construction of process data ontologies. The work of
Belhajjame and Brambilla (2009) proposes a query-
by-example approach that relies on ontological
description of business processes, activities, and
their relationships. The work of Lincoln and Gal
(2011) automatically extracts and uses the
operational layer (the “how-to”) and the business
rules encapsulated in a process repository. Such
automatic ontology extraction techniques are
important for analyzing data encapsulated in the
Web. Nevertheless, the current research literature is
based solely on process-flow structured repositories
and not on unstructured repositories such as the
Web.
KEOD2014-InternationalConferenceonKnowledgeEngineeringandOntologyDevelopment
468
3 KEY DEFICIENCIES OF
SEMANTIC BUSINESS
PROCESS SEARCH
Current Web searches are based on keyword queries
and semantic similarity lookups. This makes data
extraction relatively easy and simple for users.
Nevertheless, and as demonstrated in Section 1, it is
practically impossible to extract processes from the
Web using current semantic search engine
technology. This is an inherent material weakness
that in our opinion presents a significant barrier for
the evolution of Web usability.
The main search engines (e.g. Google, Microsoft
Bing, Ask, and others) are still at experimental,
initial phases of enabling Web process-searches. For
example, recent R&D efforts of Google yield lists of
“how to” instructions for very limited process sets.
For instance, when searching in Google “how to
issue an invoice,” a set of related documents and
media is retrieved, without any process-formatted
results. However, in some cases, we identified initial
attempts to retrieve instruction-based results for
certain “how to” queries. These results are presented
before the standard Google search results, within a
dedicated frame, in a list-format, which is a first step
in aiming to retrieve and present process-flow
formats. This presentation is still at a preliminary
phase as Google requests users' feedback regarding
the quality of these instruction lists.
Besides these attempts to present somewhat
process-driven results, we note that standard search
results for “how-to” or “process-driven” queries are
very limited, again due to the aforementioned
material weakness of semantic search. The work of
Wasser and Lincoln (2014) presents examples of
process-search scenarios using the top four search
engines (Google, Bing, Yahoo and Ask). The results
demonstrated that current search technologies cannot
provide adequate results. The sample searches
included not only business process but also personal
process queries as the authors believe that the size
and amount of data presented in the Web will also
extend the scope of process related searches beyond
the domain of BPM.
Clearly, these results encompass a large,
unstructured, set of data with a low level of
usability. Results are not presented in standardized
process notations- nor in basic flowchart formations.
Practically, for the end-user, it is not possible to
deduct an actual process from search results. It is not
clear what the required steps are and what is the
order of activities for achieving the process goal. It
is also not possible assess the quality and relevance
of the suggested results from an operational
viewpoint- as ranking is based on semantics and not
on operational characteristics of a process. Hence,
these samples along with the elaborated example
presented in section 1 demonstrate current process-
search deficiencies and the need for an alternative
framework that will support such Web searches.
The work of Wasser and Lincoln (2014)
estimates the demand for such a solution for
process-based queries. According to this work, As of
July 2014, “how-to” related searches are conducted
over 91.4 million times per year. This amount
emphasizes the need for process search, and is
expected to grow significantly when it will be
possible to retrieve proper, process-format, results
from these searches.
4 FRAMEWORK
SPECIFICATIONS FOR WEB
PROCESS SEARCHES
In this Section we describe the main expectations
from a framework for process search within
unstructured data repositories, such as the Web. So
far, this paper has focused on analyzing deficiencies
of current search method, and this section is aimed
at highlighting the high-level steps and requirements
required to fulfil an adequate framework.
The main target of the framework is to make
processes accessible and usable for everybody
through simple Web searches. The framework
should enable searches on “how to do things” using
a simple query language for expert as well as non-
expert users - using their natural language
terminology for describing the process goal. While
current Web searches are based on keywords and
semantic similarity, the suggested high-level
retrieval framework is based on operational
similarity.
The suggested framework includes an input, an
output and five processing steps (see illustration in
Fig. 2). To discuss the main concepts, we first
present the required definitions, and based on these
definitions we then describe the main requirements
for each step, as follows.
4.1 Definitions
A Business Process Model is a directed graph,
),( EVM
, where V is a set of nodes and
BusinessProcessSearchwithinUnstructuredRepositories
469
Figure 2: A high-level framework for process search on
the Web.
VVE is a set of edges. Each node v is
associated with a set of (attribute,value) pairs,
denoted
)(vA . Given an attribute a , we use )(va
to denote the set of values d s.t.
)(),( vAda . An
example of a Business Process Model (BPM) related
attributes and values is presented in Table 1. Each
edge is associated with a label, denoted
)(vl . We
use
to denote the empty label.
Table 1: An example of BPM related attributes and values
for a node, v .
Attribute (
a )
Description
Value (
d )
Type
Describes the type of data
that the node represents.
Each node posses exactly
one Type attribute
Activity, decision,
control, role, event
(including the start
and end events)
Text
Describes an action when
type=activity/decision/con
trol, and an event when
type=event. Each node
posses exactly one Text
attribute
Action: “Perform
project period
close,” “Open a
new accounting
period”
Event: “Project
termination,”
“New accounting
period started”
Role
Describes the executing
party of the node’s action
(relevant for
type=activity/decision/con
trol). Each node posses
exactly one Role attribute
Accounts
receivable clerk,
System
administrator,
Human resource
manager
Document
Represents a document
involved in the action’s
execution (relevant for
type=activity/decision/con
trol). Each node can be
related to 0-n Documents
Customers list,
pricing list, an
invoice
Figure 3: Conversion of a process flow diagram (Part a)
into the directed graph (Part b).
Fig. 3 presents a simple flowchart diagram of a
business process (Part a) and its conversion into the
directed graph,
),( EVM
, using the above
component definitions (Part b).
4.2 Input
The framework is aimed at enabling users to define
“how to” questions such as: “How can I assure
regulatory compliance in sales processes?,” “What
non-discrimination measures are being taken in HR
processes?,” and “How can I initiate a purchase
order without a purchase requisition?.” To do that, it
is required to select an adequate query language.
When searching for the most suitable query
language we should take into account two user
types: (a) the simple, common user that has no
knowledge in BPM; and (b) an expert user that is
familiar with BPM. As for the simple user - natural
language is the easiest way for phrasing his search
goal. This allows him to phrase the question in his
own words. For the expert user, the query language
should also be intuitive but will enable further
specifications that will support more specific
searches. Therefore, the query language should
enable querying the structural level as well (namely:
O
u
t
p
u
t
R
el
a
x
E
x
t
r
a
c
t
I
n
t
e
r
p
r
e
t
Type: Action
Text: Perform Projects
Period Close
Role: Stockroom Clerk
Type: Event
Text: Start
Type: Activity
Text: Perform all final runs
of system interface
processes for the period
Role: Stockroom Clerk
Type: Activity
Text: Complete all manual
invoices, credit memos, and
debit memos
Role: Accounts Receivable
Clerk
Type: Activity
Text: Complete all manual
receipts and manual receipt
corrections
Role: Accounts Receivable
Clerk
Type: Activity
Text: Ensure there are no
unprocessed adjustments
Role: Accounts Receivable
Clerk
Type:
Decision
Text:
Unprocessed
receipts?
Yes
Type: Activity
Text: Follow up with approving
Managers to resolve adjustment
issues, and obtain system
approvals
Role: Accounts Receivable Clerk
No
Type: Activity
Text: Complete or delete all
incomplete transactions
Role: Accounts Receivable Clerk
Type: Activity
Text: Perform the
Oracle Receivables sub
ledger reconciliation
Role: Stockroom Clerk
Type: Event
Text: End
Type: Event
Text: End
Yes
No
Accounts
Receivable
Clerk
Accounts
Receivable
Clerk
Accounts
Receivable
Clerk
Stockroom
Clerk
Accounts
Receivable
Clerk
Stockroom
Clerk
Accounts
Receivable
Clerk
Start
Perform all final runs
of system interface
processes for the
period
Complete all manual
invoices, credit
memos, and debit
memos
Complete all manual
receipts and manual
receipt corrections
Ensure there are no
unprocessed
adjustments
Unprocessed
receipts?
Follow up with approving
Managers to resolve
adjustment issues, and
obtain system approvals
Complete or delete all
incomplete
transactions
Perform the Oracle
Receivables sub ledger
reconciliation
Perform Projects Period
Close
End
Accounts
Receivable
Period Close
Performed
Part A: a ProcessGene representation of a process-flowPart B: a conversion of the process-flow into a directed graph, M=(V,E)
KEOD2014-InternationalConferenceonKnowledgeEngineeringandOntologyDevelopment
470
a “structure-aware query”). Taking these issues into
account, we came to the following conclusions: (a)
the query language for expert users should be based
on keywords and natural language for searching the
BPM content layer and optionally - labels for
involving also structural constraints; (b) the
combination of labels and keywords should be based
on the simple flow diagram principles.
For that, two language syntaxes should be
defined:
1. A graphical syntax – enabling the user to
express queries using flow diagram graphical
components (e.g., activities, roles, edges)
2. A textual syntax – enabling the user to express
queries using keywords/natural language phrases
within optional flow diagram patterns
More formally, the structure-aware query
language can be defined as follows. A structure-
aware query is a directed graph,
)','(' EVQ ,
where
'V is a set of nodes and ''' VVE is a set
of edges. Each node
'v is associated with a set of
(attribute, regular-expression) pairs, denoted
)'(' vA .
The regular expressions refer to the attribute values
in M’s nodes. An example for an (attribute, regular-
expression) pair is: (Text, *Financ*), which means
that the description related to this node contains the
string: “Financ”. Given an attribute
'a , we use
)'(' va to denote the set of regular expressions re
s.t.
)'('),'( vArea
. Finally, each edge is
associated with a binary function,
f , denoted
)'(vf , that returns a Boolean value. This function
describes the relationship between two nodes.
Examples of
)'(vf are:
1. A decision that leads to a report-related event.
2. Two activities that are not connected by any
path to each other.
3. Two nodes with certain attributes that are
connected by a limited path length (see illustration
in Fig. 4). According to this example, the Boolean
function associated with the edge that connects the
two nodes will returns True only if the graph path
between the nodes contains no more than two nodes.
In addition, the first node is a control action,
executed by an accountant, and the second node
represents a decision that involves a receipt-like text.
Figure 4: An example of a structure-aware query in which
two activities are connected by no more than two steps.
4.3 Process
The initial phase of the search process aims at
processing the unstructured Web data into a
structured process repository. Technically, it will not
be feasible, nor efficient, to decompose the entire
Web data into a BPM structure “on-demand” for
every search query. Therefore, this conversion
should be executed after each Web content change,
similarly to the search engine's indexing and data
mapping processes.
First, it is required to identify process “blocks”
and process elements within Web pages. We will
elaborate on this part in future works. Second, it will
be required to decompose textual phrases related to
the identified process elements (e.g. activities) into
meaningful process ontologies - in order to further
analyze their meaning and build structural process
taxonomies automatically. This should be done by
utilizing NLP methods, as well as by deploying
process textual-decomposition models. Third, based
on the extracted process structures and the generated
process ontologies, it is required to generate unique
process data taxonomies that will represent the Web
encapsulated know-how, and will further assist in
processing the search query. We will elaborate and
present examples of such taxonomies in future work.
The outcome of this phase is (a) a collection of
process-flows in the format of directed graphs; and
(b) process taxonomies that further enable the
processing of search queries on the generated
graphs.
4.4 Interpret
This phase is required only for natural language
queries. At this stage it is required to interpret the
natural language phrase into process-aware
notations. This is done automatically, and serves as a
basis for machine processing of the search query.
4.5 Extract
At this phase it is required to extract related process
segments from the structured repository that fulfil
the search goal, and combine them into process-
format search results. Note that a naïve solution at
this phase would be to examine all possible process
model segments within the repository. Nevertheless,
such algorithm can be highly inefficient. Therefore,
as a preliminary step, it is first required to reduce the
set of segments by selecting only relevant
candidates.
For natural language queries such extraction and
BusinessProcessSearchwithinUnstructuredRepositories
471
reduction process can be performed using state of
the art operational search methods. For a structure-
aware query, both goals will be achieved by
performing a graph search according to the
following definition. A query result is a mapping of
Q to
M
, in which:
For each )('' QVv , Tvf ))'((
(1)
For each )('),'( QErev
,
Trevf ))(),'((
(2)
Where
symbolizes the Boolean value True.
4.6 Relax
In cases where the search phrase retrieves only few
or no results, it is required to apply search phrase
relaxation rules to optimize the search-result range.
The location of each rule in the list will represent its
relative priority. For natural language (NL) queries,
the relaxation rules should be carried out on the
process ontology model, generated at the former
“Interpret” phase. Examples for such relaxations are
presented as follows:
1. Convert each ontology component (e.g. action,
activity, object, attribute) into their synonyms
2. Replace the action with an action located at a
higher or lower hierarchal level in the ontology
model
3. Replace the object with an object located at a
higher or lower hierarchal level in the ontology
model
4. Replace the action with a more advanced or
prior action at the ontology model
5. Replace the object with a more advanced or
prior object at the ontology model
In case of a structure-aware query, the relaxation
rules can be elaborated to express also structure-
based logics. Examples of such relaxation rules are
listed in Table 1 below, where “E” symbolizes an
edge-related relaxation rule and “N” a node-related
one.
The relaxation process ends at the earliest of
either (a) reaching the expected amount of results; or
(b) implementing all relaxation rules.
4.7 Output
The goal of this phase is to output a ranked list of
full-fledged process models. This phase includes
three main steps, as follows (see illustration in Fig.
5).
Table 1: An example of structure-aware query relaxation
rules.
# Type Example
1 E
Enlarge the limit regarding the connecting
path's length between two nodes
2 E
Enable a long path length between two
nodes, instead of a “no connecting path”
constraint
3 N
Replace “Activity” with “Decision” in all
“Type” related regular expressions, and
vice versa
4 N
Replace “Activity” with “Control” in all
“Type” related regular expressions, and
vice versa
5 N
Replace “Control” with “Decision” in all
“Type” related regular expressions, and
vice versa
6 N
Add a wild-card at the beginning and at
the end of each Text related constraint
7 N
Remove all query components related to
“Role”
8 N
Remove all query components related to
“Type”
Figure 5: The main steps comprising the “Output
preparation” phase.
4.7.1 Assess
The framework aims at returning search results
ranked by their relative importance, so that users
will be able to examine them more efficiently and
effectively. Therefore, this phase's goal is to assess
the relevance of result candidates, retrieved at the
previous phases, to the search request. In case of an
NL-based search, it is required to calculate the
proximity of each process result to the process
ontology model that represents the search query. To
do that it is possible to use proximity assessment
methods. In case of a structure-aware query it is
required to calculate the similarity between two
graphs: each result vs. the query graph. It is possible
to calculate this similarity using one of the state-of-
O
u
t
p
u
t
P
r
o
c
e
s
s
S
I
F
T
S
O
R
T
Full result list
R
e
l
e
v
a
n
c
e
s
c
o
r
e
s
KEOD2014-InternationalConferenceonKnowledgeEngineeringandOntologyDevelopment
472
the-art methods for assessing similarity between
process models, e.g. the works of Dongen et al.
(2008), Ehrig et al. (2007), and van der Aalst et al.
(2006). On top of this score, in case the advanced
structure-aware query also allows the user to specify
importance weights on each edge, it is possible to
add an additional score that reflects the weights of
the matched edges. Note that it will not be required
to handle results that were produced based on
relaxation rules differently, since they will naturally
be panelled by the similarity calculation methods.
4.7.2 Sift
At this phase several thresholds should be used to
determine the inclusion of each result candidate in
the final result list. Examples for non-inclusion rules
may be as follows: (a) very short results may be
excluded if most results are much longer; or (b)
exclusion of results in which the action flow does
not match any action sequence in the action
sequence model.
4.7.3 Sort
Eventually, it is required to sort the list of search
results according to their similarity score, as
calculated during the above “Assess” stage. As an
advanced proposition, it will also be recommended
to apply learning capabilities that will opt to
improve the ranking quality for each specific user.
An example of such learning mechanism is
presented in the work of Wasser and Lincoln (2012).
The learning mechanism in that work analyzes, in
real-time, the linguistic relationships between
process ontology models and adjusts them according
to previous human inputs. As part of the search
process it will be possible to collect such inputs from
previous searches and user-specific result selections.
The learning mechanism can increase the
effectiveness of the method.
5 CONCLUSIONS
We presented a framework for searching process
models within the Web. The framework aims to
overcome the shortcomings of existing search
technologies within unstructured repositories. The
proposed framework provides a starting point that
can already be applied in real-life scenarios, yet
several research issues remain open- to be addressed
in future research. We mention three such extensions
here. First, formalizing the framework into a detailed
executable method. Second, extending the models of
process logic for determining the ranking of
extracted results. Third, extending the set of
relaxation rules. It is hoped that by expanding search
and query capabilities of processes within the Web,
users will be able to extract operational knowledge
more simply and efficiently.
REFERENCES
Awad, A., 2007. BPMN-Q: A Language to Query
Business Processes. In EMISA, volume 119, pages
115128.
Awad, A., Polyvyanyy, A., Weske, M., 2008. Semantic
querying of business process models. In 12th
International IEEE Enterprise Distributed Object
Computing Conference, pages 8594. IEEE.
Beeri, C., Eyal, A., Kamenkovich, S., Milo, T., 2008.
Querying business processes with BP-QL. Information
Systems, 33(6):477507.
Belhajjame, K., Brambilla, M., 2009. Ontology-based
description and discovery of business processes.
Enterprise, Business-Process and Information Systems
Modeling, pages 8598.
Ehrig, M., Koschmider, A., Oberweis, A., 2007.
Measuring similarity between semantic business
process models. In Proceedings of the fourth Asia-
Pacific conference on Conceptual modelling - Volume
67, APCCM '07, pages 7180, Darlinghurst, Australia,
Australia. Australian Computer Society, Inc.
Guo, L., Shao, F., Botev, C., Shanmugasundaram, J.,
2003. XRANK: Ranked keyword search over XML
documents. In Proceedings of the 2003 ACM
SIGMOD international conference on Management of
data, pages 1627. ACM.
He, H., Wang, H., Yang, J., Yu, P.S., 2007. BLINKS:
ranked keyword searches on graphs. In Proceedings of
the 2007 ACM SIGMOD international conference on
Management of data, pages 305316. ACM.
Hristidis, V., Papakonstantinou, Y., Balmin, A., 2003.
Keyword proximity search on XML graphs.
Karni, R., Wasser, A., Lincoln, M., 2014. Content analysis
of business processes. International journal of e-
business development.
Katz, B., Lin, J., Quan, D., 2010. Natural language
annotations for the Semantic Web. On the Move to
Meaningful Internet Systems 2002: CoopIS, DOA,
and ODBASE, pages 13171331.
Leopold, H., Smirnov, S., Mendling, J., 2010. Refactoring
of process model activity labels. In Natural Language
Processing and Information Systems, pages 268276.
Springer.
Lincoln, M., Gal, A., 2011. Searching business process
repositories using operational similarity. On the Move
to Meaningful Internet Systems: OTM, pages 219.
Lincoln, M., Golani, M., Gal, A., 2010. Machine-assisted
design of business process models using descriptor
BusinessProcessSearchwithinUnstructuredRepositories
473
space analysis. Business Process Management, pages
128144.
Markovic, I., Pereira, A. C., Stojanovic, N., 2008. A
framework for querying in business process
modelling. In Proceedings of the Multikonferenz
Wirtschaftsinformatik (MKWI), Munchen, Germany.
Shao, Q., Sun, P., Chen, Y., 2009. WISE: a workflow
information search engine. In ICDE'09. IEEE 25th
International Conference on, pages 14911494. IEEE.
van der Aalst, W., de Medeiros, A., Weijters, A., 2006.
Process equivalence: Comparing two process models
based on observed behaviour. In Business Process
Management, pages 129144. Springer Berlin /
Heidelberg.
van Dongen, B. F., Dijkman, R. M., Mendling, J., 2008.
Measuring similarity between business process
models. In Advanced Information Systems
Engineering, 20th Int. Conference, CAiSE 2008,
Montpellier, France, pages 450464. Springer.
Wasser, A., Lincoln, M., 2012. Semantic machine learning
for business process content generation. In On the
Move to Meaningful Internet Systems: OTM 2012,
pages 7491. Springer.
Wasser, A., Lincoln, M., 2014. Key Deficiencies of
Semantic Business Process Search. COOPIS
workshops: INBAST.
Wasser, A., Lincoln, M., Karni, R., 2006. ProcessGene
querya tool for querying the content layer of business
process models. In Demo Session of the 4th
International Conference on Business Process
Management, pages 18.
KEOD2014-InternationalConferenceonKnowledgeEngineeringandOntologyDevelopment
474