Identification of domain information for use in machine learning models
A device may analyze a set of unstructured documents of an organization associated with a domain to identify a first set of entities. The device may analyze a set of semi-structured documents of the organization to determine a second set of entities. The device may filter the first set of entities using the second set of entities. Filtering the first set of entities may include removing, from the first set of entities, one or more entities that do not satisfy a threshold level of similarity with entities included in the second set of entities. The device may consolidate the filtered first set of entities and the second set of entities to identify a set of key entities. The device may provide the set of key entities to a user device to allow the set of key entities to be annotated and used for one or more machine learning models.
1. A device, comprising:
one or more processors to:
receive historical data of an organization associated with a particular domain,
the historical data including a set of unstructured documents and a set of semi-structured documents;
determine candidate domain information by analyzing the set of unstructured documents using one or more natural language processing techniques,
the candidate domain information including at least one of:
a first set of entities, or
a first set of associations;
determine additional candidate domain information by analyzing the set of semi-structured documents using the one or more natural language processing techniques,
the additional candidate domain information including at least one of:
a second set of entities, or
a second set of associations, and
where the one or more processors, when determining the additional candidate domain information, are to:
determine the second set of entities by parsing the set of semi-structured documents with document-specific metadata;
filter the candidate domain information using the additional candidate domain information,
where the one or more processors, when filtering candidate domain information, are to:
compare an entity, of the first set of entities, and one or more entities of the second set of entities,
determine that the entity, of the first set of entities, does not satisfy a threshold level of similarity with the one or more entities of the second set of entities, and
remove the entity, of the first set of entities, based on determining that the entity does not satisfy the threshold level of similarity with the one or more entities of the second set of entities;
filter the additional candidate domain information using the candidate domain information;
consolidate the filtered candidate domain information and the filtered additional candidate domain information into key domain information;
generate a data structure that associates values included in the key domain information; and
provide the key domain information to a user device to allow the key domain information to be annotated and used for one or more machine learning models.
2. The device of claim 1 , where the set of unstructured documents and/or the set of semi-structured documents include documents with text in different languages.
3. The device of claim 1 , where the one or more processors, when determining the candidate domain information, are to:
determine the first set of entities by analyzing the set of unstructured documents using an entity identification technique,
the entity identification technique to parse through the set of unstructured documents to create the first set of entities; and
where the one or more processors, when determining the additional candidate domain information, are to:
analyze the set of semi-structured documents to identify the document-specific metadata included in the set of semi-structured documents,
the document-specific metadata including at least one of:
information associated with a document layout,
information associated with positioning of entities within a document, or
information associated with format indicators.
4. The device of claim 1 , where the one or more processors, when determining the candidate domain information, are to:
determine the first set of associations by analyzing the set of unstructured documents using a first association detection technique,
the first association detection technique to use lexical indicators and sentence context information to determine the first set of associations; and
where the one or more processors, when determining the additional candidate domain information, are to:
determine the second set of associations by analyzing the set of semi-structured documents using a second association detection technique,
the second association detection technique to use the document-specific metadata included in the set of semi-structured documents to determine the second set of associations.
5. The device of claim 1 , where the one or more processors, when filtering the additional candidate domain information, are to:
compare an association, of the second set of associations, and one or more associations of the first set of associations,
determine that the association, of the second set of associations, does not satisfy a threshold level of similarity with the one or more associations of the first set of associations, and
remove the association, of the second set of associations, based on determining that the association does not satisfy the threshold level of similarity with the one or more associations of the first set of associations.
6. The device of claim 1 , where the one or more processors, when providing the key domain information to the user device, are to:
provide the key domain information for display on a user interface of the user device,
the key domain information including a set of key entities and a set of key associations,
where the user interface displays the set of key entities and the set of key associations in a manner that allows the set of key entities and the set of key associations to be annotated and used for a machine learning model of the one or more machine learning models.
7. The device of claim 1 , where the one or more processors are further to:
execute a testing procedure to test an accuracy of the key domain information.
8. A method, comprising:
receiving, by a device, historical data of an organization associated with a particular domain,
the historical data including a set of unstructured documents and a set of semi-structured documents;
analyzing, by the device, the set of unstructured documents using one or more natural language processing techniques to identify a first set of entities;
determining, by the device, a second set of entities by analyzing the set of semi-structured documents using the one or more natural language processing techniques,
where determining the second set of entities comprises:
determining the second set of entities by parsing the set of semi-structured documents with document-specific metadata;
filtering, by the device, the first set of entities using the second set of entities,
where filtering the first set of entities includes:
comparing one or more entities, of the first set of entities, and one or more entities of the second set of entities,
determining that the one or more entities, of the first set of entities, does not satisfy a threshold level of similarity with the one or more entities of the second set of entities, and
removing, from the first set of entities, the one or more entities that do not satisfy a threshold level of similarity with the one or more entities included in the second set of entities;
consolidating, by the device, the filtered first set of entities and the second set of entities to identify a set of key entities;
generating, by the device, a data structure that is used to store the set of key entities; and
providing, by the device, the set of key entities to a user device to allow the set of key entities to be annotated and used for one or more machine learning models.
9. The method of claim 8 , further comprising:
determining, after receiving the historical data, a first set of associations associated with the set of unstructured documents;
determining, after receiving the historical data, a second set of associations associated with the set of semi-structured documents;
filtering the second set of associations using the first set of associations,
where filtering the second set of associations includes removing, from the second set of associations, one or more associations that do not satisfy a threshold level of similarity with associations included in the first set of associations;
consolidating the first set of associations and the filtered second set of associations to identify a set of key associations; and
where generating the data structure comprises:
generating the data structure,
the data structure to associate the set of key entities with the set of key associations.
10. The method of claim 8 , further comprising:
filtering, after analyzing the set of semi-structured documents, the second set of entities using the first set of entities,
where the filtering of the second set of entities includes removing, from the second set of entities, one or more entities that do not satisfy a threshold level of similarity with entities included in the first set of entities.
11. The method of claim 8 , where the set of unstructured documents and/or the set of semi-structured documents include text written in different languages.
12. The method of claim 8 , where determining the first set of entities comprises:
analyzing the set of unstructured documents using an entity identification technique,
the entity identification technique to parse through the set of unstructured documents to determine the first set of entities, and
sorting the first set of entities using an entity frequency-inverse entity frequency (EF-IDF) technique; and
where determining the second set of entities comprises:
analyzing the set of semi-structured documents to identify the document-specific metadata included in the set of semi-structured documents, and
determining the second set of entities by parsing the set of semi-structured documents using the document-specific metadata.
13. The method of claim 8 , where providing the set of key entities to the user device comprises:
providing the set of key entities for display on a user interface of the user device,
where the user interface displays the set of key entities in a manner that allows the set of key entities to be annotated and used for a machine learning model of the one or more machine learning models.
14. The method of claim 8 , further comprising:
executing a testing procedure to test an accuracy of the set of key entities.
15. A non-transitory computer-readable medium storing instructions, the instructions comprising:
one or more instructions that, when executed by one or more processors, cause the one or more processors to:
receive historical data of an organization associated with a particular domain,
the historical data including a set of unstructured documents and a set of semi-structured documents;
determine candidate domain information by analyzing the set of unstructured documents using one or more natural language processing techniques,
the candidate domain information including at least one of:
a first set of entities, or
a first set of values;
determine additional candidate domain information by analyzing the set of semi-structured documents using the one or more natural language processing techniques,
the additional candidate domain information including at least one of:
a second set of entities, or
a second set of values, and
where the one or more instructions, that cause the one or more processors to determine the additional candidate domain information, cause the one or more processors to:
determine the additional candidate domain information by parsing the set of semi-structured documents with document-specific metadata;
filter the candidate domain information using the additional candidate domain information,
where the one or more processors, when filtering candidate domain information, are to:
compare an entity, of the first set of entities, and one or more entities of the second set of entities,
determine that the entity, of the first set of entities, does not satisfy a threshold level of similarity with the one or more entities of the second set of entities, and
remove the entity, of the first set of entities, based on determining that the entity does not satisfy the threshold level of similarity with the one or more entities of the second set of entities;
filter the additional candidate domain information using the candidate domain information;
consolidate the filtered candidate domain information and the filtered additional candidate domain information into key domain information;
generate a data structure that associates values included in the key domain information; and
provide the key domain information to a user device to allow the key domain information to be annotated and used for one or more machine learning models.
16. The non-transitory computer-readable medium of claim 15 , where the one or more instructions, when executed by the one or more processors, further cause the one or more processors to:
determine, after receiving the historical data, a first set of associations associated with the set of unstructured documents,
where the first set of associations is part of the candidate domain information;
determine, after receiving the historical data, a second set of associations associated with the set of semi-structured documents,
where the second set of associations is part of the additional candidate domain information;
filter the second set of associations using the first set of associations; and
consolidate the first set of associations and the filtered second set of associations to identify a set of key associations,
where the set of key associations is part of the key domain information.
17. The non-transitory computer-readable medium of claim 15 , where the one or more instructions, that cause the one or more processors to receive the historical data, cause the one or more processors to:
receive a first unstructured document, of the set of unstructured documents, with text written in a first language,
receive a second unstructured document, of the set of unstructured documents with text written in a second language,
receive a first semi-structured document, of the set of semi-structured documents, with text written in a first language, and
receive a second semi-structured document, of the set of semi-structured documents, with text written in a second language.
18. The non-transitory computer-readable medium of claim 15 , where the one or more instructions, that cause the one or more processors to determine the candidate domain information, cause the one or more processors to:
analyze the set of unstructured documents using an entity identification technique,
the entity identification technique to parse through the set of unstructured documents to determine the first set of entities,
sort the first set of entities using a ranking technique; and
where the one or more instructions, that cause the one or more processors to determine the additional candidate domain information, cause the one or more processors to:
analyze the set of semi-structured documents to identify document-specific metadata included in the set of semi-structured documents, and
determine the second set of entities by parsing the set of semi-structured documents using the document-specific metadata.
19. The non-transitory computer-readable medium of claim 15 , where the one or more instructions, that cause the one or more processors to filter the candidate domain information using the additional candidate domain information, cause the one or more processors to:
filter the first set of values by removing, from the first set of values, one or more values that do not satisfy a threshold level of similarity with values included in the second set of values.
20. The non-transitory computer-readable medium of claim 15 , where the one or more instructions, that cause the one or more processors to provide the key domain information to the user device, cause the one or more processors to:
provide the key domain information for display on a user interface of the user device,
the key domain information including a set of key entities and a set of key values,
receive, from the user device, a machine learning model, of the one or more machine learning models, that includes an annotated set of key entities and an annotated set of key values,
receive additional data associated with the organization, and
provide the additional data as input to the machine learning model to cause the machine learning model to output one or more predictions associated with improving one or more processes within the organization.