Document analysis architecture
Systems and methods for generation and use of document analysis architectures are disclosed. A model builder component may be utilized to receiving user input data for labeling a set of documents as in class or out of class. That user input data may be utilized to train one or more classification models, which may then be utilized to predict classification of other documents. Trained models may be incorporated into a model taxonomy for searching and use by other users for document analysis purposes.
1. A system comprising:
one or more processors; and
non-transitory computer-readable media storing computer-executable instructions that, when executed by the one or more processors, cause the one or more processors to perform operations comprising:
receiving, via a user interface associated with a platform for classifying documents as in class or out of class, first user input data indicating a first portion of the documents are in class;
receiving, via the user interface, second user input data indicating a second portion of the documents are out of class;
determining a first set of features of the first portion of the documents that are representative of the first portion of the documents;
determining a second set of features of the second portion of the documents that are representative of the second portion of the documents;
training a classification model based at least in part on the first features and the second features, the classification model trained to determine a classification of individual ones of the documents by analyzing features of the individual ones of the documents in association with the first features and the second features;
determining, utilizing the classification model, a third portion of the documents that are in class;
determining, utilizing the classification model, a fourth portion of the documents that are out of class;
receiving third user input data indicating additional classification determinations associated with the documents other than the first portion of the documents;
retraining the classification model based at least in part on the third user input data;
determining a difference in the number of the documents in the third portion after utilizing the classification model as retrained; and
generating a labeling influence value indicating a degree of influence on the classification model by additional user input data.
2. The system of claim 1 , the operations further comprising:
determining a first confidence score that a document of the documents correlates to the first features;
determining a second confidence score that the document correlates to the second features;
determining that the first confidence score indicates more confidence than the second confidence score; and
associating the document with the third portion of the documents based at least in part on the first confidence score indicating more confidence than the second confidence score.
3. The system of claim 1 , the operations further comprising:
determining a document type of the documents, the document type indicating at least one of a subject matter associated with the documents, a database from which the documents were received, or a format of the documents; and
selecting a base model of multiple models based at least in part on the document type, wherein the base model has been configured to generate output utilizing the document type.
4. A method, comprising:
receiving first user input data indicating a first portion of documents are in class;
receiving second user input data indicating a second portion of the documents are out of class;
determining a first set of features of the first portion of the documents that are representative of the first portion of the documents;
determining a second set of features of the second portion of the documents that are representative of the second portion of the documents;
training a model based at least in part on the first features and the second features, the model trained to determine a classification of individual ones of the documents;
determining, based at least in part on the model, a third portion of the documents that are in class;
determining, based at least in part on the model, a fourth portion of the documents that are out of class;
receiving third user input data indicating additional classification determinations associated with the documents other than the first portion of the documents;
retraining the model based at least in part on the third user input data;
determining a difference in the number of the documents in the third portion after utilizing the model as retrained; and
generating a labeling influence value indicating a degree of influence on the model by additional user input data.
5. The method of claim 4 , further comprising:
determining that a document of the documents correlates to the first features more than the second features; and
associating the document with the third portion of the documents based at least in part on the document correlating to the first features more than the second features.
6. The method of claim 4 , further comprising:
determining a document type of the documents; and
selecting a base model of multiple models based at least in part on the document type, wherein the base model has been configured to generate output utilizing the document type.
7. The method of claim 4 , further comprising:
determining, for individual ones of the first portion of the documents, a first vector representing the first features;
determining, for individual ones of the second portion of the documents, a second vector representing the second features; and
determining that a document of the documents is associated with the third portion of the documents based at least in part on a third vector representing the document being associated with the first vector more than the second vector.
8. The method of claim 4 , further comprising:
identifying additional documents that differ from the documents;
determining a classification of individual ones of the additional documents utilizing the model;
determining a confidence value indicating performance of the model for determining the classification of the individual ones of the additional documents; and
determining that the model has been trained successfully based at least in part on the confidence value satisfying a threshold confidence value.
9. The method of claim 4 , further comprising:
determining first classification codes associated with the first portion of the documents;
determining second classification codes associated with the second portion of the documents;
determining, utilizing the model, that a first sample document is to be classified as in class based at least in part on the first sample document including keywords representing the first sample document and the first sample document being associated with at least one of the first classification codes; and
determining, utilizing the model, that a second sample document is to be classified as out of class based at least in part on the second sample document including the keywords and the second sample document being associated with at least one of the second classification codes.
10. The method of claim 4 , further comprising:
receiving third user input data indicating classification of a document associated with the third portion of the documents, the third user input data indicating that the document is out of class;
determining third features representing the document; and
retraining the model based at least in part on the third features.
11. A system comprising:
one or more processors; and
non-transitory computer-readable media storing computer-executable instructions that, when executed by the one or more processors, cause the one or more processors to perform operations comprising:
receiving first user input data indicating a first portion of documents are in class;
receiving second user input data indicating a second portion of the documents are out of class;
determining a first set of features of the first portion of the documents that are representative of the first portion of the documents;
determining a second set of features of the second portion of the documents that are representative of the second portion of the documents;
training a model based at least in part on the first features and the second features, the model trained to determine a classification of individual ones of the documents;
determining, based at least in part on the model, a third portion of the documents that are in class;
determining, based at least in part on the model, a fourth portion of the documents that are out of class;
receiving third user input data indicating additional classification determinations associated with the documents other than the first portion of the documents;
retraining the model based at least in part on the third user input data;
determining a difference in the number of the documents in the third portion after utilizing the model as retrained; and
generating a labeling influence value indicating a degree of influence on the model by additional user input data.
12. The system of claim 11 , the operations further comprising:
determining that a document of the documents correlates to the first features more than the second features; and
associating the document with the third portion of the documents based at least in part on the document correlating to the first features more than the second features.
13. The system of claim 11 , the operations further comprising:
determining a document type of the documents; and
selecting a base model of multiple models based at least in part on the document type, wherein the base model has been configured to generate output utilizing the document type.
14. The system of claim 11 , the operations further comprising:
determining, for individual ones of the first portion of the documents, a first vector representing the first features;
determining, for individual ones of the second portion of the documents, a second vector representing the second features; and
determining that a document of the documents is associated with the third portion of the documents based at least in part on a third vector representing the document being associated with the first vector more than the second vector.
15. The system of claim 11 , the operations further comprising:
identifying additional documents that differ from the documents;
determining a classification of individual ones of the additional documents utilizing the model;
determining a confidence value indicating performance of the model for determining the classification of the individual ones of the additional documents; and
determining that the model has been trained successfully based at least in part on the confidence value satisfying a threshold confidence value.
16. The system of claim 11 , the operations further comprising:
causing display of keywords representing the first features via a user interface;
receiving user input, via the user interface, indicating that a set of documents associated with a keyword of the keywords should be marked as in class instead of out of class; and
retraining the model based at least in part on the user input.
17. The system of claim 11 , the operations further comprising:
receiving third user input data indicating classification of a document associated with the third portion of the documents, the third user input data indicating that the document is out of class;
determining third features representing the document; and
retraining the model based at least in part on the third features.