Document classification and characterization using human judgment, tiered similarity analysis and language/concept analysis
Systems, methods, and articles are provided for characterizing and defining groups within large corpuses of documents using a combination of one or more of human judgment, tiered similarity analysis techniques, and language/concept analysis. Related apparatus, systems, techniques and articles are also described.
1. A method comprising:
receiving a corpus of documents;
characterizing similarities among the corpus of documents using at least three similarity algorithms having different similarity criteria, the characterizing comprising:
obtaining contextual characteristics for each of the corpus of documents and associating the contextual characteristics with the corresponding document, the contextual characteristics selected from a group consisting of: similarity score, type of similarity algorithm used to characterize the document, document family, document type, and metadata describing properties of the document;
first removing a first portion of the corpus of documents based on applying a first similarity algorithm to the corpus of documents;
second removing, after the first removing, a second portion of the corpus of documents based on applying a second similarity algorithm to the corpus of documents; and
third removing, after the first removing and the second removing, a third portion of the corpus of documents based on applying a third similarity algorithm to the corpus of documents, the third similarity algorithm based on a criteria other than that implemented by the first similarity algorithm and the second similarity algorithm, wherein the third similarity algorithm identifies conceptually similar documents in the corpus of documents based on content of each respective document, and wherein the conceptually similar documents are neither exact duplicates nor substantial duplicates;
defining stacks of documents based on pre-defined grouping criteria as applied to the characterized similarities among the corpus of documents, the characterized similarities based on the first removing, the second removing, or the third removing;
identifying, within each stack, a prime document; and
initiating provision of each prime document to at least one human reviewer via a computer-implemented document review and characterization system.
2. The method as in claim 1 , wherein at least one of the similarity algorithms identifies exact duplicates in the corpus of documents.
3. The method as in claim 1 , wherein at least one of the similarity algorithms identifies substantial duplicates in the corpus of documents.
4. The method as in claim 1 , wherein the pre-defined grouping criteria is adjusted based on quality control review of documents within the stack other than the corresponding prime document.
5. The method as in claim 1 , further comprising: receiving data characterizing quality control review of at least a portion of the corpus of documents, the received data being used to modify the pre-defined grouping criteria to either increase or decrease one or more similarity metrics.
6. The method as in claim 1 , wherein at least a portion of the corpus of documents comprise families of documents, the families of documents having a pre-defined logical interrelation.
7. A method comprising:
receiving a corpus of documents;
obtaining contextual characteristics for each of the corpus of documents and associating the contextual characteristics with the corresponding document, the contextual characteristics selected from a group consisting of: similarity score, type of similarity algorithm used to characterize the document, document family, document type, and metadata describing properties of the document;
generating a first subset of the corpus of documents by identifying and characterizing similarities among the corpus of documents based on applying a first similarity algorithm to the corpus of documents;
generating a second subset of the corpus of documents by identifying and characterizing similarities among the first subset of the corpus of documents based on applying a second similarity algorithm to the first subset of the corpus of documents, the second similarity algorithm having a relaxed similarity standard as compared to the first similarity algorithm; and
generating a third subset of the corpus of documents by identifying and characterizing similarities among the corpus of documents based on applying a third similarity algorithm to the second subset of the corpus of documents, the third similarity algorithm having a similarity standard as other than that implemented by the first similarity algorithm and the second similarity algorithm, wherein the third similarity algorithm identifies conceptually similar documents in the corpus of documents based on content of each respective document, and wherein the conceptually similar documents are neither exact duplicates nor substantial duplicates;
defining stacks of documents based on pre-defined grouping criteria as applied to the second subset of the corpus of documents and the third subset of the corpus of documents;
identifying, within each stack, a prime document; and
initiating provision of each prime document to at least one human reviewer via a computer-implemented document review and characterization system.
8. A non-transitory computer program product storing instructions, which when executed by at least one data processor of at least one computing system, result in operations comprising:
receiving a corpus of documents;
characterizing similarities among the corpus of documents using at least three similarity algorithms having different similarity criteria, the characterizing comprising:
obtaining contextual characteristics for each of the corpus of documents and associating the contextual characteristics with the corresponding document, the contextual characteristics selected from a group consisting of: similarity score, type of similarity algorithm used to characterize the document, document family, document type, and metadata describing properties of the document;
first removing a first portion of the corpus of documents based on applying a first similarity algorithm to the corpus of documents; and
second removing, after the first removing, a second portion of the corpus of documents based on applying a second similarity algorithm to the corpus of documents; and
third removing, after the first removing and the second removing, a third portion of the corpus of documents based on applying a third similarity algorithm to the corpus of documents, the third similarity algorithm based on a criteria other than that implemented by the first similarity algorithm and the second similarity algorithm, wherein the third similarity algorithm identifies conceptually similar documents in the corpus of documents based on content of each respective document, and wherein the conceptually similar documents are neither exact duplicates nor substantial duplicates;
defining stacks of documents based on pre-defined grouping criteria as applied to the characterized similarities among the corpus of documents, the characterized similarities based on the first removing, the second removing, or the third removing;
identifying, within each stack, a prime document; and
initiating provision of each prime document to at least one human reviewer via a computer-implemented document review and characterization system.
9. The non-transitory computer program product as in claim 8 , wherein at least one of the similarity algorithms identifies exact duplicates in the corpus of documents.
10. The non-transitory computer program product as in claim 8 , wherein at least one of the similarity algorithms identifies substantial duplicates in the corpus of documents.
11. The non-transitory computer program product as in claim 8 , wherein the pre-defined grouping criteria is adjusted based on quality control review of documents within the stack other than the corresponding prime document.
12. The non-transitory computer program product as in claim 8 , wherein the operations further comprise: receiving data characterizing quality control review of at least a portion of the corpus of documents, the received data being used to modify the pre-defined grouping criteria to either increase or decrease one or more similarity metrics.
13. The non-transitory computer program product as in claim 8 , wherein at least a portion of the corpus of documents comprise families of documents, the families of documents having a pre-defined logical interrelation.
14. A system comprising:
at least one data processor; and
memory storing instructions, which when executed by the at least one data processor, result in operations comprising:
receiving a corpus of documents;
characterizing similarities among the corpus of documents using at least three similarity algorithms having different similarity criteria, the characterizing comprising:
obtaining contextual characteristics for each of the corpus of documents and associating the contextual characteristics with the corresponding document, the contextual characteristics selected from a group consisting of: similarity score, type of similarity algorithm used to characterize the document, document family, document type, and metadata describing properties of the document;
first removing a first portion of the corpus of documents based on applying a first similarity algorithm to the corpus of documents;
second removing, after the first removing, a second portion of the corpus of documents based on applying a second similarity algorithm to the corpus of documents; and
third removing, after the first removing and the second removing, a third portion of the corpus of documents based on applying a third similarity algorithm to the corpus of documents, the third similarity algorithm based on a criteria other than that implemented by the first similarity algorithm and the second similarity algorithm, wherein the third similarity algorithm identifies conceptually similar documents in the corpus of documents based on content of each respective document, and wherein the conceptually similar documents are neither exact duplicates nor substantial duplicates;
defining stacks of documents based on pre-defined grouping criteria as applied to the characterized similarities among the corpus of documents, the characterized similarities based on the first removing, the second removing, or the third removing;
identifying, within each stack, a prime document; and
initiating provision of each prime document to at least one human reviewer via a computer-implemented document review and characterization system.