Methods and systems for processing documents with task-specific highlighting
Methods and systems for automatically processing a document may include classifying a document, such as a medical document, as one or more document types based at least in part on one or more machine learning models and one or more tokens extracted from the medical document, determining a token contribution weight of each token towards the classification, modifying the medical document based on the token contribution weights of the one or more tokens, and displaying the modified medical document on a display to a user.
1. A computer-implemented method comprising:
generating a classification of a medical document as one or more document types based at least in part on one or more machine learning models and one or more tokens extracted from the medical document;
determining a contribution weight of a phrase or a sentence in the medical document toward the classification based on a sum of token contribution weights of tokens in the phrase or the sentence;
modifying the medical document, wherein the phrase or the sentence is modified to reflect a degree of visual emphasis based on the contribution weight of the phrase or the sentence; and
causing display of the modified medical document.
2. The computer-implemented method of claim 1 , wherein generating the classification comprises classifying the medical document as including text relating to a particular medical event.
3. The computer-implemented method of claim 2 , wherein the medical event is a clinical exam.
4. The computer-implemented method of claim 1 , wherein generating the classification comprises:
determining at least one bag of words vector from one or more extracted tokens;
generating at least one topic vector from the at least one bag of words vector, wherein the at least one topic vector comprises one or more topic features, each topic feature mapping to a probability distribution of tokens for the topic feature; and
classifying the medical document based on the at least one topic vector.
5. The computer-implemented method of claim 4 , wherein generating at least one topic vector comprises fitting a Latent Dirichlet Allocation model to the at least one bag of words vector.
6. The computer-implemented method of claim 4 , wherein classifying the medical document comprises applying a gradient boosting classifier model to the at least one topic vector.
7. The computer-implemented method of claim 4 , further comprising:
determining an impact score associated with each topic feature in the at least one topic vector;
distributing the impact scores to the one or more tokens according to the probability distributions mapped to each topic feature in the at least one topic vector; and
determining a token contribution weight for each token by summing the distributed impact scores for the token.
8. The computer-implemented method of claim 7 , wherein the impact score is a Shapley Additive Explanation (SHAP) value, and wherein distributing the impact scores to the one or more tokens comprises distributing the SHAP value of each topic feature to all of the one or more tokens, and normalizing the distributed SHAP values.
9. The computer-implemented method of claim 1 , wherein modifying the medical document comprises mapping the contribution weight to a color map, and modifying text of the medical document based on the mapping.
10. The computer-implemented method of claim 9 , wherein modifying the medical document comprises at least one of:
changing at least one of text font appearance and text font highlighting based on the mapping; and
scaling the color map based on a confidence level of the classification of the medical document.
11. The computer-implemented method of claim 1 , further comprising determining a HEDIS quality metric based at least in part on the classification of the medical document.
12. A system comprising:
one or more processors; and
memory storing one or more instructions that, when executed by the one or more processors, cause the system to perform operations comprising:
generating a classification of a medical document as one or more document types based at least in part on one or more machine learning models and one or more tokens extracted from the medical document;
determining a contribution weight of a phrase or a sentence in the medical document toward the classification based on a sum of token contribution weights of tokens in the phrase or the sentence;
modifying the medical document, wherein the phrase or the sentence is modified to reflect a degree of visual emphasis based on the contribution weight of the phrase or the sentence; and
causing display of the modified medical document.
13. The system of claim 12 , wherein generating the classification comprises:
determining at least one bag of words vector from one or more extracted tokens;
generating at least one topic vector from the at least one bag of words vector, wherein the at least one topic vector comprises one or more topic features, each topic feature mapping to a probability distribution of tokens for the topic feature; and
classifying the medical document based on the at least one topic vector.
14. The system of claim 13 , wherein the operations further comprise:
determining an impact score associated with each topic feature in the at least one topic vector;
distributing the impact scores to the one or more tokens according to the probability distributions mapped to each topic feature in the at least one topic vector; and
determining a token contribution weight for each token by summing the distributed impact scores for the token.
15. The system of claim 12 , wherein modifying the medical document comprises mapping the contribution weight to a color map, and at least one of modifying text of the medical document based on the mapping, and changing at least one of text font appearance and text font highlighting based on the mapping.
16. A non-transitory computer-readable storage medium including instructions that, when executed, cause a computing system to perform operations comprising:
generating a classification of a medical document as one or more document types based at least in part on one or more machine learning models and one or more tokens extracted from the medical document;
determining a token contribution weight of a phrase or a sentence in the medical document toward the classification based on a sum of token contribution weights of tokens in the phrase or the sentence;
modifying the medical document, wherein the phrase or the sentence is modified to reflect a degree of visual emphasis based on the contribution weight of the phrase or the sentence; and
causing display of the modified medical document on a display.
17. The non-transitory computer-readable storage medium of claim 16 , wherein the operations further comprise:
determining an impact score associated with each topic feature in the at least one topic vector;
distributing the impact scores to the one or more tokens according to the probability distributions mapped to each topic feature in the at least one topic vector; and
determining a token contribution weight for each token by summing the distributed impact scores for the token.
18. The non-transitory computer-readable storage medium of claim 16 , wherein modifying the medical document comprises mapping the contribution weight to a color map, and at least one of modifying text of the medical document based on the mapping, and changing at least one of text font appearance and text font highlighting based on the mapping.
19. The system of claim 12 , wherein generating the classification comprises classifying the medical document as including text relating to a particular medical event, and the medical event is a clinical exam.
20. The non-transitory computer-readable storage medium of claim 16 , wherein generating the classification comprises classifying the medical document as including text relating to a particular medical event, and the medical event is a clinical exam.