Machine learning document classification

1. A method comprising:

receiving a message including an attachment document, wherein the message relates to a real estate transaction;

determining that the attachment document has a document type corresponding to one or more particular document types;

determining one or more first features from content of the attachment document;

providing the first features to one or more classification models, the one or more classification models including a machine learning model, wherein the machine learning model is trained to generate a prediction of one or more classifications of attachment documents, each classification corresponding to a type of form associated with a real estate transaction, the generating the prediction being based on the one or more first features determined from the content of the attachment document and a plurality of model parameters trained based on a collection of messages, each having one or more labeled attachments corresponding to particular types of forms, wherein the collection of messages are associated with a plurality of real estate transactions;

generating one or more predicted classifications of the attachment document; and

associating the one or more predicted classifications with the attachment document.

2. The method of claim 1 , wherein determining one or more first features from content of the attachment document comprises:

standardizing text content of the attachment document; and

generating the one or more first features from vectors computed from n-grams of the standardized text content.

3. The method of claim 2 , wherein determining one or more first features further comprises performing text recognition on the attachment document, wherein the text recognition is associated with one or more confidence values.

4. The method of claim 3 , wherein the one or more classification models includes a rule based model, wherein the rule based model applies one or more rules to the input features, and wherein the rule-based model is applied to the input features based on the one or more confidence values.

5. The method of claim 4 , wherein in response to determining that a classification output by the machine learning model is different from a classification output by the rule-based model, the classification of the rule-based model is set as the predicted classification of the attachment document.

6. The method of claim 1 , further comprising:

determining one or more second features from content of the message, wherein determining one or more second features from content of the message comprises:

generating one or more features based on text of the message matching predefined words or phrases, wherein the matching satisfies a specified degree of match between the text of the message and the predefined words or phrases.

7. The method of claim 6 , wherein determining one or more second features from the content or the message comprises:

determining the presence of a unique identifier in the content of the message corresponding to a unique file record; and

generating one or more features based on the content of the file record.

8. One or more non-transitory computer-readable storage media encoded with instructions that, when executed by one or more computers, cause the one or more computers to perform operations comprising:

receiving a message including an attachment document, wherein the message relates to a real estate transaction;

determining that the attachment document has a document type corresponding to one or more particular document types;

determining one or more first features from content of the attachment document;

providing the first features to one or more classification models, the one or more classification models including a machine learning model, wherein the machine learning model is trained to generate a prediction of one or more classifications of attachment documents, each classification corresponding to a type of form associated with a real estate transaction, the generating the prediction being based on the one or more first features determined from the content of the attachment document and a plurality of model parameters trained based on a collection of messages, each having one or more labeled attachments corresponding to particular types of forms, wherein the collection of messages are associated with a plurality of real estate transactions;

generating one or more predicted classifications of the attachment document; and

associating the one or more predicted classifications with the attachment document.

9. The non-transitory computer-readable storage media of claim 8 , wherein determining one or more first features from content of the attachment document comprises:

standardizing text content of the attachment document; and

generating the one or more first features from vectors computed from n-grams of the standardized text content.

10. The non-transitory computer-readable storage media of claim 9 , wherein determining one or more first features further comprises performing text recognition on the attachment document, wherein the text recognition is associated with one or more confidence values.

11. The non-transitory computer-readable storage media of claim 10 , wherein the one or more classification models includes a rule based model, wherein the rule based model applies one or more rules to the input features, and wherein the rule-based model is applied to the input features based on the one or more confidence values.

12. The non-transitory computer-readable storage media of claim 11 , wherein in response to determining that a classification output by the machine learning model is different from a classification output by the rule-based model, the classification of the rule-based model is set as the predicted classification of the attachment document.

13. The non-transitory computer-readable storage media of claim 8 , further comprising:

determining one or more second features from content of the message, wherein determining one or more second features from content of the message comprises:

generating one or more features based on text of the message matching predefined words or phrases, wherein the matching satisfies a specified degree of match between the text of the message and the predefined words or phrases.

14. The non-transitory computer-readable storage media of claim 13 , wherein determining one or more second features from the content or the message comprises:

determining the presence of a unique identifier in the content of the message corresponding to a unique file record; and

generating one or more features based on the content of the file record.

15. A system comprising:

one or more computers and one or more storage devices storing instructions that are operable, when executed by the one or more computers, to cause the one or more computers to perform operations comprising:

receiving a message including an attachment document, wherein the message relates to a real estate transaction;

determining that the attachment document has a document type corresponding to one or more particular document types;

determining one or more first features from content of the attachment document;

providing the first features to one or more classification models, the one or more classification models including a machine learning model, wherein the machine learning model is trained to generate a prediction of one or more classifications of attachment documents, each classification corresponding to a type of form associated with a real estate transaction, the generating the prediction being based on the one or more first features determined from the content of the attachment document and a plurality of model parameters trained based on a collection of messages, each having one or more labeled attachments corresponding to particular types of forms, wherein the collection of messages are associated with a plurality of real estate transactions;

generating one or more predicted classifications of the attachment document; and

associating the one or more predicted classifications with the attachment document.

16. The system of claim 15 , wherein determining one or more first features from content of the attachment document comprises:

standardizing text content of the attachment document; and

generating the one or more first features from vectors computed from n-grams of the standardized text content.

17. The system of claim 16 , wherein determining one or more first features further comprises performing text recognition on the attachment document, wherein the text recognition is associated with one or more confidence values.

18. The system of claim 17 , wherein the one or more classification models includes a rule based model, wherein the rule based model applies one or more rules to the input features, and wherein the rule-based model is applied to the input features based on the one or more confidence values.

19. The system of claim 18 , wherein in response to determining that a classification output by the machine learning model is different from a classification output by the rule-based model, the classification of the rule-based model is set as the predicted classification of the attachment document.

20. The system of claim 15 , further comprising:

determining one or more second features from content of the message, wherein determining one or more second features from content of the message comprises:

generating one or more features based on text of the message matching predefined words or phrases, wherein the matching satisfies a specified degree of match between the text of the message and the predefined words or phrases.

21. The system of claim 20 , wherein determining one or more second features from the content or the message comprises:

determining the presence of a unique identifier in the content of the message corresponding to a unique file record; and

generating one or more features based on the content of the file record.