Citation Context

...velop strategy, whose business object strategy occurs in six categories of the PCF taxonomy. In such a situation, it would be helpful to quantify the similarity between the entire process model and the different taxonomy categories. The resulting values could complement the individual similarity scores derived from sim in order to truly identify the best fitting candidate. To quantify the similarity between a process model and a taxonomy category, we adopt the general idea of term frequency-inverse document frequency (tf-idf) and the vector space model from the domain of information retrieval [23] and modify them to meet the characteristics of our problem. Using the tf-idf it is possible to determine the discriminative power of a word. In the context of a 6 Leopold et al. taxonomy, the tf-idf assigns high weights to words that frequently occur in a particular category but rarely in the entire taxonomy. Contrarily, words that occur in many or even all categories have hardly any discriminative power and hence receive a low weight. Let Ac denote the concepts from a category c ∈ C and f(w, c) the frequency of a word w among the concepts of c. Then, the tf-idf is defined as follows: tf -idf...

Citation Context

... of concepts on the path from a2 to as, and N3 is the number of concepts on the path from as to the taxonomy root r. By calculating dc for each concept pair, we obtain a set of distance cost values in the interval [-1,1], i.e., only big leaps are penalized. Together with the Towards the Automated Annotation of Process Models 7 previously introduced similarity values they form the input for our probabilistic annotation model. 4 Using Markov Logic for Automatic Annotation Markov Logic (ML) is a formalism that combines first order logic with undirected probabilistic models, i.e., Markov Networks [21]. A Markov Logic formalization of a given problem consists of a set of weighted and unweighted logical formulae. These formulae describe observations relevant to the concrete problem instance and general constraints that have to hold for each instance of the problem class. By replacing variables with concrete values, which is called grounding, it is possible to transform the Markov Logic formalization into a Markov Logic Network. For technical details we refer to [21]. Two types of inference can be applied to a Markov Logic Network, known as marginal inference and maximum a-posteriori (MAP) in...

Citation Context

...stance Costs between Taxonomy Concepts Besides the semantic perspective, we also need to take the structure of the process model and the taxonomy into account. Assuming that process models describe the underlying process in a rather coherent fashion, we would not expect large “leaps” between the annotations of two neighboring activities. For instance, we would assume that two subsequent activities are rather annotated with concepts 6.2.1 and 6.2.6 than with 7.1 and 3.2.1.5. To penalize such leaps, we introduce a distance cost function dc based on concept similarity introduced by Wu and Palmer [28]. It quantifies the distance between two concepts based on the graph structure of the taxonomy and their least common superconcept as. Given two concepts a1, a2 ∈ At, we define dc as follows: dc(a1, a2) = ( − 0.5 + 2×N3 N1 +N2 + 2×N3 ) × 2 (6) where N1 is the number of concepts on the path from a1 to as, N2 is the number of concepts on the path from a2 to as, and N3 is the number of concepts on the path from as to the taxonomy root r. By calculating dc for each concept pair, we obtain a set of distance cost values in the interval [-1,1], i.e., only big leaps are penalized. Together with the To...

Citation Context

...omy consisting of 1,131 concepts demonstrates that our approach produces satisfying results. Keywords: Process Model, Taxonomy, Automatic Annotation 1 Introduction Nowadays, many organizations use business process models for documenting and improving their operations. However, only a few have recognized the full potential their process models offer. In particular semantic technologies facilitate a wide range of possibilities that go beyond the documentation of business operations [27]. For example, there are techniques available that use process models for checking business process compliance [9,22], for checking the interoperability of business processes [10], and for discovering semantic weaknesses in business processes [2]. However, the limitation of all these approaches is that they build on an existing annotation of the process model activities, for instance, with concepts from a taxonomy. Recognizing this drawback, user-friendly approaches for semantic annotation have been proposed [4]. Still, the manual effort that is required for annotating process models is considerable and, in many cases, even 2 Leopold et al. hardly manageable taking into account that taxonomies often contain ...

Citation Context

...ognizing this drawback, user-friendly approaches for semantic annotation have been proposed [4]. Still, the manual effort that is required for annotating process models is considerable and, in many cases, even 2 Leopold et al. hardly manageable taking into account that taxonomies often contain hundreds or even thousands of concepts. In this paper, we present the first approach for automatically annotating process models with the concepts of a taxonomy. At this stage, we focus on activity-based taxonomies such as the Supply-Chain Operations Reference-model (SCOR) [24], the MIT process handbook [17], and the Process Classification Framework (PCF) [1]. To this end, we define an approach that combines semantic similarity measurement with probabilistic optimization. In particular, we use different types of similarity between the process model and the taxonomy as well as the distance between the taxonomy concepts to guide the matching with a Markov Logic formalization. In contrast to prior approaches in the domain of process modeling, we do not measure the similarity using WordNet, but build on the more powerful corpus-based approach of second-order similarity. An evaluation of our approach ...

Citation Context

...terview, and the additional information fragment with employee. The same procedure can be applied to the concepts of an activity taxonomy. In order to accomplish this decomposition in an automated way, we employ the technique defined in [15]. Building on the decomposition, we compute the semantic similarity between the actions, business objects, and additional information fragments of the considered activity-concept pair. A challenge in this context is the usage of specific terminology from business settings, which is often not fully captured by standard natural language tools such as WordNet [19]. Hence, we determine the similarity between two components using a corpus-based method called second-order similarity [11]. The approach of second-order similarity is based on the statistical analysis of co-occurrences in large text collections and has been implemented in several tools such as NLS [5] or DISCO [13]. In comparison to WordNet, second-order similarity has the advantage that it is not restricted to a set of manually predefined term relations and, hence, is more powerful for our purposes. In order to calculate the semantic similarity between a process model activity ap Towards the...

Citation Context

...omy consisting of 1,131 concepts demonstrates that our approach produces satisfying results. Keywords: Process Model, Taxonomy, Automatic Annotation 1 Introduction Nowadays, many organizations use business process models for documenting and improving their operations. However, only a few have recognized the full potential their process models offer. In particular semantic technologies facilitate a wide range of possibilities that go beyond the documentation of business operations [27]. For example, there are techniques available that use process models for checking business process compliance [9,22], for checking the interoperability of business processes [10], and for discovering semantic weaknesses in business processes [2]. However, the limitation of all these approaches is that they build on an existing annotation of the process model activities, for instance, with concepts from a taxonomy. Recognizing this drawback, user-friendly approaches for semantic annotation have been proposed [4]. Still, the manual effort that is required for annotating process models is considerable and, in many cases, even 2 Leopold et al. hardly manageable taking into account that taxonomies often contain ...

Citation Context

...gy [4,7,3]. The final decision about the annotation is, however, still taken by the user. Hence, we are, to the best of our knowledge, the first who present an automatic approach for annotating process models. Process model matching aims at the automatic identification of correspondences between two process models. In prior work, a plethora of process model matching approaches has been proposed [6]. Typically, they build on a combination of structural or behavioral properties with different types of textual similarity. Some rely on rather simplistic techniques such as the Levenshtein distance [26], others use WordNet for computing textual similarity [14,12]. However, so far, no approach has considered the use of second-order similarity. Besides these conceptual differences, it is worth noting that the overall complexity of automated annotation is considerably higher. While a process model typically does not consist of more than 30 activities, taxonomies often contain more than thousand concepts [1]. 14 Leopold et al. 7 Conclusion In this paper, we presented the first approach for automatically annotating process models with concepts of a taxonomy. Our approach uses a Markov Logic forma...

Citation Context

...sion and recall metrics for our context is that they only consider annotations that are correct up to the last sub category. As an example, consider an activity ap that was manually annotated with the concept 6.2.1.4. If our approach proposes to annotate ap with the concept 6.1.2.2, this would be simply considered as incorrect although the first three levels of the manually annotated concept were actually identified correctly. To provide for a more fine granular perspective on our results, we introduce a level-based form of precision and, recall which is in line with the approach presented in [8]. To this end, we introduce a function parenti(at), which returns the i th parent of a concept at ∈ At. We define parenti(at) = parent(parenti−1) for all i > 0 and parenti(at) = at for all i ≤ 0. Thus, for instance, parent2(at) returns the concept 6.2 for the input concept 6.2.1.4. We further introduce a function level(at), which returns the level of a concept at in the taxonomy. It, for example, returns 4 for the concept 6.2.1.4 and 2 for the concept 6.2. Based on these definitions, we introduce a function ln, which maps a set of annotations A to a set of less fine-grained annotations from le...

Citation Context

...s pointed out in [18], activities can be characterized by three components: an action, a business object on which the action is performed, and an optional additional information fragment that is providing further details. As an example, consider the process model activity Perform interview with employee. It consists of the action perform, the business object interview, and the additional information fragment with employee. The same procedure can be applied to the concepts of an activity taxonomy. In order to accomplish this decomposition in an automated way, we employ the technique defined in [15]. Building on the decomposition, we compute the semantic similarity between the actions, business objects, and additional information fragments of the considered activity-concept pair. A challenge in this context is the usage of specific terminology from business settings, which is often not fully captured by standard natural language tools such as WordNet [19]. Hence, we determine the similarity between two components using a corpus-based method called second-order similarity [11]. The approach of second-order similarity is based on the statistical analysis of co-occurrences in large text col...

Citation Context

...size as well as their coverage of the different PCF categories. Moreover, some models are cross sectional (i.e., models 5, 6, and 7) while others only belong to a single PCF category. Thus, we believe that the test set is well-suited to demonstrate the applicability of our annotation approach. 10 Leopold et al. 5.2 Setup For evaluating the approach presented in this paper, we implemented it in the context of a prototype. The prototype is based on the activity analysis technique from [15], the second-order similarity implementation DISCO [13], and the Markov Logic Network implementation RockIt [20]. We used our prototype to automatically generate the annotations for the models from our test set and the text-based PCF taxonomy version 5.2. We then compared the automatically generated annotations A with the manual annotation from the students R. Based on this comparison, we can assess the quality of the annotation by computing the metrics precision and recall: pre(A,R) = |A ∩ R| |A| rec(A,R) = |A ∩ R| |R| In our context, precision is the number of correct annotations computed by our approach divided by the number of annotations our approach proposed. Recall is the number of correct annota...

Citation Context

...produces satisfying results. Keywords: Process Model, Taxonomy, Automatic Annotation 1 Introduction Nowadays, many organizations use business process models for documenting and improving their operations. However, only a few have recognized the full potential their process models offer. In particular semantic technologies facilitate a wide range of possibilities that go beyond the documentation of business operations [27]. For example, there are techniques available that use process models for checking business process compliance [9,22], for checking the interoperability of business processes [10], and for discovering semantic weaknesses in business processes [2]. However, the limitation of all these approaches is that they build on an existing annotation of the process model activities, for instance, with concepts from a taxonomy. Recognizing this drawback, user-friendly approaches for semantic annotation have been proposed [4]. Still, the manual effort that is required for annotating process models is considerable and, in many cases, even 2 Leopold et al. hardly manageable taking into account that taxonomies often contain hundreds or even thousands of concepts. In this paper, we pres...

Citation Context

...de range of possibilities that go beyond the documentation of business operations [27]. For example, there are techniques available that use process models for checking business process compliance [9,22], for checking the interoperability of business processes [10], and for discovering semantic weaknesses in business processes [2]. However, the limitation of all these approaches is that they build on an existing annotation of the process model activities, for instance, with concepts from a taxonomy. Recognizing this drawback, user-friendly approaches for semantic annotation have been proposed [4]. Still, the manual effort that is required for annotating process models is considerable and, in many cases, even 2 Leopold et al. hardly manageable taking into account that taxonomies often contain hundreds or even thousands of concepts. In this paper, we present the first approach for automatically annotating process models with the concepts of a taxonomy. At this stage, we focus on activity-based taxonomies such as the Supply-Chain Operations Reference-model (SCOR) [24], the MIT process handbook [17], and the Process Classification Framework (PCF) [1]. To this end, we define an approach th...

Citation Context

...rity between the actions, business objects, and additional information fragments of the considered activity-concept pair. A challenge in this context is the usage of specific terminology from business settings, which is often not fully captured by standard natural language tools such as WordNet [19]. Hence, we determine the similarity between two components using a corpus-based method called second-order similarity [11]. The approach of second-order similarity is based on the statistical analysis of co-occurrences in large text collections and has been implemented in several tools such as NLS [5] or DISCO [13]. In comparison to WordNet, second-order similarity has the advantage that it is not restricted to a set of manually predefined term relations and, hence, is more powerful for our purposes. In order to calculate the semantic similarity between a process model activity ap Towards the Automated Annotation of Process Models 5 and an taxonomy concept at, we introduce three functions: a component similarity function cpsim, a coverage function cov, and a activity-concept similarity function sim, combining the latter two into a final result. The function cpsim calculates the semantic si...

Citation Context

...the actions, business objects, and additional information fragments of the considered activity-concept pair. A challenge in this context is the usage of specific terminology from business settings, which is often not fully captured by standard natural language tools such as WordNet [19]. Hence, we determine the similarity between two components using a corpus-based method called second-order similarity [11]. The approach of second-order similarity is based on the statistical analysis of co-occurrences in large text collections and has been implemented in several tools such as NLS [5] or DISCO [13]. In comparison to WordNet, second-order similarity has the advantage that it is not restricted to a set of manually predefined term relations and, hence, is more powerful for our purposes. In order to calculate the semantic similarity between a process model activity ap Towards the Automated Annotation of Process Models 5 and an taxonomy concept at, we introduce three functions: a component similarity function cpsim, a coverage function cov, and a activity-concept similarity function sim, combining the latter two into a final result. The function cpsim calculates the semantic similarity betwe...

Citation Context

...ivity taxonomy. In order to accomplish this decomposition in an automated way, we employ the technique defined in [15]. Building on the decomposition, we compute the semantic similarity between the actions, business objects, and additional information fragments of the considered activity-concept pair. A challenge in this context is the usage of specific terminology from business settings, which is often not fully captured by standard natural language tools such as WordNet [19]. Hence, we determine the similarity between two components using a corpus-based method called second-order similarity [11]. The approach of second-order similarity is based on the statistical analysis of co-occurrences in large text collections and has been implemented in several tools such as NLS [5] or DISCO [13]. In comparison to WordNet, second-order similarity has the advantage that it is not restricted to a set of manually predefined term relations and, hence, is more powerful for our purposes. In order to calculate the semantic similarity between a process model activity ap Towards the Automated Annotation of Process Models 5 and an taxonomy concept at, we introduce three functions: a component similarity ...

Citation Context

...v Logic formalization. An evaluation with a set of 12 process models consisting of 148 activities and the PCF taxonomy consisting of 1,131 concepts demonstrates that our approach produces satisfying results. Keywords: Process Model, Taxonomy, Automatic Annotation 1 Introduction Nowadays, many organizations use business process models for documenting and improving their operations. However, only a few have recognized the full potential their process models offer. In particular semantic technologies facilitate a wide range of possibilities that go beyond the documentation of business operations [27]. For example, there are techniques available that use process models for checking business process compliance [9,22], for checking the interoperability of business processes [10], and for discovering semantic weaknesses in business processes [2]. However, the limitation of all these approaches is that they build on an existing annotation of the process model activities, for instance, with concepts from a taxonomy. Recognizing this drawback, user-friendly approaches for semantic annotation have been proposed [4]. Still, the manual effort that is required for annotating process models is consid...

Citation Context

...igned it to 6.2.3.4 Select and reject candidates. Although the manual annotation is an arguably better choice than the computed one, both concepts represent semantically similar choices. 6 Related Work The work presented in this paper relates to two major streams of research: process model annotation and process model matching. Research addressing process model annotation typically aims at describing general guidelines and strategies [16] or the benefits and potentials associated with the annotation [25]. Some approaches also automatically filter relevant concepts from the considered ontology [4,7,3]. The final decision about the annotation is, however, still taken by the user. Hence, we are, to the best of our knowledge, the first who present an automatic approach for annotating process models. Process model matching aims at the automatic identification of correspondences between two process models. In prior work, a plethora of process model matching approaches has been proposed [6]. Typically, they build on a combination of structural or behavioral properties with different types of textual similarity. Some rely on rather simplistic techniques such as the Levenshtein distance [26], othe...

Citation Context

...h concepts from a taxonomy. Recognizing this drawback, user-friendly approaches for semantic annotation have been proposed [4]. Still, the manual effort that is required for annotating process models is considerable and, in many cases, even 2 Leopold et al. hardly manageable taking into account that taxonomies often contain hundreds or even thousands of concepts. In this paper, we present the first approach for automatically annotating process models with the concepts of a taxonomy. At this stage, we focus on activity-based taxonomies such as the Supply-Chain Operations Reference-model (SCOR) [24], the MIT process handbook [17], and the Process Classification Framework (PCF) [1]. To this end, we define an approach that combines semantic similarity measurement with probabilistic optimization. In particular, we use different types of similarity between the process model and the taxonomy as well as the distance between the taxonomy concepts to guide the matching with a Markov Logic formalization. In contrast to prior approaches in the domain of process modeling, we do not measure the similarity using WordNet, but build on the more powerful corpus-based approach of second-order similarity....

Citation Context

...test set candidate from model 9. Our approach annotated this activity with the concept 6.2.4.3 Recommend/not recommend candidate whereas the manual annotation assigned it to 6.2.3.4 Select and reject candidates. Although the manual annotation is an arguably better choice than the computed one, both concepts represent semantically similar choices. 6 Related Work The work presented in this paper relates to two major streams of research: process model annotation and process model matching. Research addressing process model annotation typically aims at describing general guidelines and strategies [16] or the benefits and potentials associated with the annotation [25]. Some approaches also automatically filter relevant concepts from the considered ontology [4,7,3]. The final decision about the annotation is, however, still taken by the user. Hence, we are, to the best of our knowledge, the first who present an automatic approach for annotating process models. Process model matching aims at the automatic identification of correspondences between two process models. In prior work, a plethora of process model matching approaches has been proposed [6]. Typically, they build on a combination of ...

Citation Context

...wever, still taken by the user. Hence, we are, to the best of our knowledge, the first who present an automatic approach for annotating process models. Process model matching aims at the automatic identification of correspondences between two process models. In prior work, a plethora of process model matching approaches has been proposed [6]. Typically, they build on a combination of structural or behavioral properties with different types of textual similarity. Some rely on rather simplistic techniques such as the Levenshtein distance [26], others use WordNet for computing textual similarity [14,12]. However, so far, no approach has considered the use of second-order similarity. Besides these conceptual differences, it is worth noting that the overall complexity of automated annotation is considerably higher. While a process model typically does not consist of more than 30 activities, taxonomies often contain more than thousand concepts [1]. 14 Leopold et al. 7 Conclusion In this paper, we presented the first approach for automatically annotating process models with concepts of a taxonomy. Our approach uses a Markov Logic formalization to combine the results of different similarity funct...

Citation Context

...wever, still taken by the user. Hence, we are, to the best of our knowledge, the first who present an automatic approach for annotating process models. Process model matching aims at the automatic identification of correspondences between two process models. In prior work, a plethora of process model matching approaches has been proposed [6]. Typically, they build on a combination of structural or behavioral properties with different types of textual similarity. Some rely on rather simplistic techniques such as the Levenshtein distance [26], others use WordNet for computing textual similarity [14,12]. However, so far, no approach has considered the use of second-order similarity. Besides these conceptual differences, it is worth noting that the overall complexity of automated annotation is considerably higher. While a process model typically does not consist of more than 30 activities, taxonomies often contain more than thousand concepts [1]. 14 Leopold et al. 7 Conclusion In this paper, we presented the first approach for automatically annotating process models with concepts of a taxonomy. Our approach uses a Markov Logic formalization to combine the results of different similarity funct...

Citation Context

...matic Annotation 1 Introduction Nowadays, many organizations use business process models for documenting and improving their operations. However, only a few have recognized the full potential their process models offer. In particular semantic technologies facilitate a wide range of possibilities that go beyond the documentation of business operations [27]. For example, there are techniques available that use process models for checking business process compliance [9,22], for checking the interoperability of business processes [10], and for discovering semantic weaknesses in business processes [2]. However, the limitation of all these approaches is that they build on an existing annotation of the process model activities, for instance, with concepts from a taxonomy. Recognizing this drawback, user-friendly approaches for semantic annotation have been proposed [4]. Still, the manual effort that is required for annotating process models is considerable and, in many cases, even 2 Leopold et al. hardly manageable taking into account that taxonomies often contain hundreds or even thousands of concepts. In this paper, we present the first approach for automatically annotating process models ...

Citation Context

...scribing general guidelines and strategies [16] or the benefits and potentials associated with the annotation [25]. Some approaches also automatically filter relevant concepts from the considered ontology [4,7,3]. The final decision about the annotation is, however, still taken by the user. Hence, we are, to the best of our knowledge, the first who present an automatic approach for annotating process models. Process model matching aims at the automatic identification of correspondences between two process models. In prior work, a plethora of process model matching approaches has been proposed [6]. Typically, they build on a combination of structural or behavioral properties with different types of textual similarity. Some rely on rather simplistic techniques such as the Levenshtein distance [26], others use WordNet for computing textual similarity [14,12]. However, so far, no approach has considered the use of second-order similarity. Besides these conceptual differences, it is worth noting that the overall complexity of automated annotation is considerably higher. While a process model typically does not consist of more than 30 activities, taxonomies often contain more than thousand ...

Citation Context

...igned it to 6.2.3.4 Select and reject candidates. Although the manual annotation is an arguably better choice than the computed one, both concepts represent semantically similar choices. 6 Related Work The work presented in this paper relates to two major streams of research: process model annotation and process model matching. Research addressing process model annotation typically aims at describing general guidelines and strategies [16] or the benefits and potentials associated with the annotation [25]. Some approaches also automatically filter relevant concepts from the considered ontology [4,7,3]. The final decision about the annotation is, however, still taken by the user. Hence, we are, to the best of our knowledge, the first who present an automatic approach for annotating process models. Process model matching aims at the automatic identification of correspondences between two process models. In prior work, a plethora of process model matching approaches has been proposed [6]. Typically, they build on a combination of structural or behavioral properties with different types of textual similarity. Some rely on rather simplistic techniques such as the Levenshtein distance [26], othe...