Automated recognition of process modeling semantics in flow diagrams
An example embodiment disclosed is a system for automated model extraction of documents containing flow diagrams. An extractor is configured to extract from the flow diagrams flow graphs. The extractor further extracts nodes and edges, and relational, geometric and textual features for the extracted nodes and edges. A classifier is configured to recognize process semantics based on the extracted nodes and edges, and the relational, geometric and textual features of the extracted nodes and edges. A process modeling language code is generated based on the recognized process semantics. Rules to recognize patterns in process diagrams may be determined using supervised learning and/or unsupervised learning. During supervised learning, an expert labels example flow diagrams so that a classifier can derive the classification rules. During unsupervised learning flow diagrams are clustered based on relational, geometric and textual features of nodes and edges.
1. A method comprising:
receiving one or more documents that contain a first flow diagram in one or more diagram formats supported by the documents;
automatically extracting from the first flow diagram one or more flow graphs comprising extracted nodes and edges;
automatically extracting from the first flow diagram relational, geometric and textual features for the extracted nodes and edges;
automatically learning rules to recognize process semantics based on the extracted nodes and edges, and the relational, geometric and textual features of the extracted nodes and edges, the rules configured as a decision tree; and
automatically generating, based on the learned rules, process modeling recognition code to recognize and decide process semantics in a second flow diagram.
2. The method of claim 1 , wherein recognition of process semantics is further based on a measured similarity of the relational, geometric and textual features of the extracted nodes and edges.
3. The method of claim 1 , wherein the textual features include a lexical category of a label associated with an extracted node or edge from the extracted nodes and edges.
4. The method of claim 1 , wherein the relational features include at least one of the number of incoming extracted edges and the number of outgoing extracted edges associated with an extracted node or edge from the extracted nodes and edges.
5. The method of claim 1 , wherein the geometric features include at least one of the shape, the number of vertical lines, the number of horizontal lines, the number of arcs, the width, the height, and the line style associated with an extracted node or edge from the extracted nodes and edges.
6. The method of claim 1 , further comprising training a classifier to identify patterns in features of the extracted nodes and edges that indicate a class of process semantic of the respective extracted nodes and edges.
7. The method of claim 6 , wherein training the classifier includes receiving by the classifier a plurality of example flow diagrams.
8. The method of claim 1 , further comprising generating the decision tree for identifying the process semantics.
9. A system comprising:
a computer processor;
computer readable storage medium coupled to the computer processor, the computer readable storage medium including one or more documents containing a first flow diagram in one or more diagram formats supported by the documents;
an extractor configured to extract from the first flow diagram one or more flow graphs comprising extracted nodes and edges, and extract from the first flow diagram relational, geometric and textual features for the extracted nodes and edges;
a classifier trained to learn rules to recognize process semantics based on the relational, geometric and textual features of the extracted nodes and edges, the rules configured as a decision tree; and
generated, based on the learned rules, process modeling recognition code to recognize and decide process semantics in a second flow diagram.
10. The system of claim 9 , wherein the classifier is configured to recognize process semantics based on a measured similarity of the relational, geometric and textual features of the extracted nodes and edges.
11. The system of claim 9 , wherein the textual features include a lexical category of a label associated with an extracted node or edge from the extracted nodes and edges.
12. The system of claim 9 , wherein the relational features include at least one of the number of incoming extracted edges and the number of outgoing extracted edges associated with an extracted node or edge from the extracted nodes and edges.
13. The system of claim 9 , wherein the geometric features include at least one of the shape, the number of vertical lines, the number of horizontal lines, the number of arcs, the width, the height, and the line style associated with an extracted node or edge from the extracted nodes and edges.
14. The system of claim 9 , wherein the classifier is trained to identify patterns in features of the extracted nodes and edges that indicate a class of process semantic of the respective extracted nodes and edges.
15. The system of claim 14 , further comprising a plurality of example flow diagrams to train the classifier.
16. The system of claim 9 , further comprising a decision tree generated by the classifier for identifying the process semantics.
17. A computer program product comprising:
a non-transitory computer readable storage medium having computer readable program code embodied therewith, the computer readable program code configured to:
receive one or more documents that contain a first flow diagram in one or more diagram formats supported by the documents;
automatically extract from the first flow diagram one or more flow graphs comprising extracted nodes and edges;
automatically extract from the first flow diagram relational, geometric and textual features for the extracted nodes and edges;
automatically learning rules to recognize process semantics based on the relational, geometric and textual features of the extracted nodes and edges, the rules configured as a decision tree; and
automatically generate, based on the learned rules, process modeling recognition code to recognize and decide process semantics in a second flow diagram.
18. The computer program product of claim 17 , wherein recognition of process semantics is further based on a measured similarity of the relational, geometric and textual features of the extracted nodes and edges.
19. The computer program product of claim 17 , wherein the textual features include a lexical category of a label associated with an extracted node or edge from the extracted nodes and edges.
20. The computer program product of claim 17 , wherein the relational features include at least one of the number of incoming extracted edges and the number of outgoing extracted edges associated with an extracted node or edge from the extracted nodes and edges.
21. The computer program product of claim 17 , wherein the geometric features include at least one of the shape, the number of vertical lines, the number of horizontal lines, the number of arcs, the width, the height, and the line style associated with an extracted node or edge from the extracted nodes and edges.
22. The computer program product of claim 17 , further comprising computer readable program code configured to train a classifier to identify patterns in features of the extracted nodes and edges that indicate a class of process semantic of the respective extracted nodes and edges.
23. The computer program product of claim 22 , further comprising computer readable program code configured to receive by the classifier a plurality of example flow diagrams.
24. The computer program product of claim 17 , wherein the computer readable program code configured to automatically generate process modeling recognition code based on the recognized process semantics includes computer readable program code configured to automatically generate the process modeling recognition code based on the learned rules includes forming learned rules based on a number of incoming and outgoing edges, and a number of vertical and horizontal lines of the extracted nodes and edges.
25. The method of claim 1 , wherein automatically generating the process modeling recognition code based on the learned rules includes forming learned rules based on a number of incoming and outgoing edges, and a number of vertical and horizontal lines of the extracted nodes and edges.