Graph similarity and alignment determination
A computing system and methods for aligning process graph representations and processing multi-representational datasets are disclosed. A first process graph representation is aligned with a second reference process graph representation using a matcher implemented in a matcher code module. A similarity evaluation is performed at one or more levels of process abstraction, and process-wide metrics are generated to assess overall alignment quality. A user interface renders the metric results and allows user modification of alignment parameters or element correspondences. Additionally, a dataset with at least two representations is processed to generate embedding vectors using different embedding techniques. The embedding vectors are combined into a hybrid vector representation, which is analyzed to determine a similarity measure relative to an input query. Search results are rendered based on the similarity measure. The disclosed techniques improve alignment accuracy, computational efficiency, and the relevance of search results in multi-representational datasets.
1 . A computing system comprising:
at least one hardware processor;
at least one memory coupled to the at least one hardware processor; and
one or more computer-readable storage media comprising computer-executable instructions that, when executed, cause the computing system to perform operations comprising:
receiving a first process graph representation, the first process graph representation defining a plurality of process elements and relationships between the plurality of process elements;
receiving a second process graph representation, the second process graph representation defining a plurality of reference process elements and relationships between the plurality of reference process elements;
determining an alignment between the first process graph representation and the second process graph representation with at least one matcher implemented in a first matcher code module, wherein the alignment associates process elements of the first process graph representation with reference process elements of the second process graph representation based on a similarity evaluation, the similarity evaluation being performed at one or more levels of process abstraction;
analyzing the alignment using at least one process-wide metric implemented in a first metric code module to generate first metric results, wherein the at least one process-wide metric is distinct from similarity evaluations performed at an element level and is computed based on aggregated characteristics of the determined alignment across multiple process elements, wherein the at least one process-wide metric evaluates the alignment across an entirety of the first process graph representation and the second process graph representation, rather than individual element correspondences;
rendering the first metric results for display via a first user interface, the first metric results including at least a process-wide alignment score indicative of an overall similarity between the first process graph representation and the second process graph representation; and;
(1) modifying at least one alignment parameter based on first user interaction with the displayed first metric results; or
(2) modifying an alignment between an element of the first process graph representation and an element of the second process graph representation based in second user interaction with the displayed first metric results.
2 . The computing system of claim 1 , wherein the similarity evaluation is performed at multiple levels of process abstraction.
3 . The computing system of claim 1 , the operations further comprising:
rendering a second user interface displaying a plurality of process-wide metrics, respective process-wide metrics of the plurality of process-wide metrics being associated with a respective metric code module;
receiving through the second user interface a selection of the at least one process-wide metric from the plurality of process-wide metrics; and
configuring a metric analysis process to use the first metric code module in analyzing the alignment.
4 . The computing system of claim 1 , the operations further comprising:
rendering a second user interface displaying a plurality of matcher techniques, respective matcher techniques of the plurality of matcher techniques being associated with a respective matcher code module;
receiving through the second user interface a selection of the at least one matcher code module; and
configuring a metric analysis process to use the at least one matcher code module in analyzing the alignment.
5 . The computing system of claim 1 , wherein the first user interaction comprises receiving user input accepting or rejecting a processed match between all or a portion of the first process graph representation and all or a portion of the second process graph representation.
6 . The computing system of claim 1 , wherein determining an alignment comprises:
generating a first embedding vector for at least one of the first process graph representation and the second process graph representation, or a component thereof, using a first embedding technique;
generating a second embedding vector for a same process graph representation, or a component thereof, using a second embedding technique different from the first embedding technique;
combining the first embedding vector and the second embedding vector to produce a combined embedding vector; and
using the combined embedding vector, or a modified version thereof, to analyze alignment between the first process graph representation and the second process graph representation.
7 . The computing system of claim 6 , the operations further comprising:
generating the first embedding vector for the first process graph representation or the second process graph representation by encoding a textual representation of a respective graph representation, or a component thereof, using a first embedding technique that processes natural language input; and
generating the second embedding vector by encoding a structured representation of the respective process graph representation, or a component thereof, using a second embedding technique that processes graph-based input.
8 . The computing system of claim 6 , wherein:
generating the first embedding vector based on a process element of the first process graph representation or the second process graph representation; and
generating the second embedding vector based on a parent node of the process element, the parent node representing a higher-level structural context for the process element within the respective process graph representation.
9 . The computing system of claim 8 , wherein:
the parent node represents a process category node that corresponds to a classification of the process element within an overarching process model.
10 . The computing system of claim 8 , wherein:
the parent node represents a containing process node that defines a structured relationship between the process element and a corresponding process workflow.
11 . The computing system of claim 8 , wherein:
the parent node represents a schema classification node that defines an ontological categorization of the process element within a structured schema model.
12 . The computing system of claim 8 , wherein:
the parent node represents a functional reference node that defines a structural or operational dependency between the process element and a corresponding reference model component.
13 . The computing system of claim 1 , wherein:
the similarity evaluation is performed at one or more levels selected from:
a process-level evaluation, wherein process workflows are compared based on execution structure and dependencies;
an activity-level evaluation, wherein individual process elements are compared based on functional similarity;
a schema-level evaluation, wherein process metadata is compared to determine alignment of structured attributes; or
a functional reference-level evaluation, wherein process elements are compared against predefined reference structures.
14 . One or more computer-readable storage media comprising:
computer-executable instructions that, when executed by a computing system comprising at least one hardware processor and at least one memory coupled to the at least one hardware process, cause the computing system to receive a dataset comprising a plurality of data elements, the dataset having at least a first representation and a second representation, wherein the first representation conveys a first type of semantic information about the dataset, and the second representation conveys a second type of semantic information about the dataset that differs from the first type;
computer-executable instructions that, when executed by the computing system, cause the computing system to generate a first embedding vector from the first representation using a first embedding technique;
computer-executable instructions that, when executed by the computing system, cause the computing system to generate a second embedding vector from the second representation using a second embedding technique different from the first embedding technique;
computer-executable instructions that, when executed by the computing system, cause the computing system to generate a unified embedding vector based on both the first embedding vector and the second embedding vector;
computer-executable instructions that, when executed by the computing system, cause the computing system to process the unified embedding vector to determine a similarity measure between the dataset and an input embedding vector corresponding to a search query;
analyzing an alignment between the first data set and a second data set, provided as search criteria, using the unified embedding vector, or a modified version thereof, and an embedding vector for the second data set; and
computer-executable instructions that, when executed by the computing system, cause the computing system to return search results in response to the search query based on the similarity measure.
15 . The one or more computer-readable storage media of claim 14 , wherein:
(1) the first type of semantic information corresponds to descriptive or contextual attributes of the dataset; and
(2) the second type of semantic information corresponds to relational or structural characteristics of the dataset.
16 . The one or more computer-readable storage media of claim 14 , wherein:
(1) the first embedding vector is generated from a natural language text representation of the dataset using a neural language model; and
(2) the second embedding vector is generated from a structured graph representation of the dataset using a graph embedding technique.
17 . The one or more computer-readable storage media of claim 14 , wherein:
(1) generating the second embedding vector further comprises selecting a parent node corresponding to a data element within the dataset, the parent node representing a higher-level structural or relational context for the data element within the dataset; and
(2) the second embedding vector is generated based on the parent node.
18 . The one or more computer-readable storage media of claim 14 , further comprising:
computer-executable instructions that, when executed by the computing system, cause the computing system to apply a dimensionality reduction technique to the unified embedding vector to generate a reduced embedding vector prior to determining the similarity measure, wherein the dimensionality reduction technique preserves key distinguishing features while improving computational efficiency.
19 . The one or more computer-readable storage media of claim 14 , wherein the search query used to determine the similarity measure comprises at least one of:
(1) a single data element;
(2) a subset of data elements from another dataset; or
(3) an entire dataset representation.
20 . A method, implemented in a computing system comprising at least one hardware processor and at least one memory coupled to the at least one hardware processor, the method comprising:
receiving a first process graph representation, the first process graph representation defining a plurality of process elements and relationships between the plurality of process elements;
receiving a second process graph representation, the second process graph representation defining a plurality of reference process elements and relationships between the plurality of reference process elements;
determining an alignment between the first process graph representation and the second process graph representation with at least one matcher implemented in a first matcher code module, wherein the alignment associates process elements of the first process graph representation with reference process elements of the second process graph representation based on a similarity evaluation, the similarity evaluation being performed at one or more levels of process abstraction;
analyzing the alignment using at least one process-wide metric implemented in a first metric code module to generate first metric results, wherein the at least one process-wide metric is distinct from similarity evaluations performed at an element level and is computed based on aggregated characteristics of the determined alignment across multiple process elements, wherein the at least one process-wide metric evaluates the alignment across an entirety of the first process graph representation and the second process graph representation, rather than individual element correspondences;
rendering the first metric results for display via a first user interface, the first metric results including at least a process-wide alignment score indicative of an overall similarity between the first process graph representation and the second process graph representation; and
(1) modifying at least one alignment parameter based on first user interaction with the displayed first metric results; or
(2) modifying an alignment between an element of the first process graph representation and an element of the second process graph representation based in second user interaction with the displayed first metric results.