Using serial machine learning models to extract data from electronic documents
Methods, systems, and apparatus, including computer programs encoded on computer storage media, for machine learning. One of the methods includes receiving a document having a plurality of first text strings; extracting the plurality of first text strings from the document; providing the extracted plurality of first text strings to a first machine learning model, wherein the first machine learning model is trained to output a numerical vector representation for each input first text string; providing the output vector representations from the first machine learning model to a second machine learning model, wherein the second machine learning model is trained to output a second text string for each input vector representation; and processing the second text strings to generate an output.
1. A method comprising:
receiving a document having a plurality of first text strings each representing a respective description related to a real estate transaction and a control value associated with the real estate transaction and having a first value from which values associated with one or more descriptions of the plurality of first text strings depend;
extracting the plurality of first text strings and the control value from the document;
validating the values associated with the respective descriptions of the plurality of first text strings in the document, the validating comprising comparing the control value to a corresponding stored value;
in response to determining a match between the control value and the stored value, providing the extracted plurality of first text strings to a first machine learning model, wherein the first machine learning model is trained to output a numerical vector representation for each input first text string, wherein each vector representation represents a meaning of the corresponding description;
providing the output vector representations from the first machine learning model to a second machine learning model, wherein the second machine learning model is trained to output a second text string representing a respective title production system description related to the real estate transaction for each input vector representation such that for each description corresponding to an input first text string a corresponding title production system description is output; and
processing the second text strings to generate an output.
2. The method of claim 1 , wherein the first machine learning model is a document to vector model trained using a collection of text strings to generate vectors that represent the meaning of the text strings.
3. The method of claim 2 , wherein the second machine learning model is a classification model trained using a collection of vectors generated by the first machine learning model to generate a prediction of a particular type of a set of types.
4. The method of claim 1 , wherein generating an output comprises:
comparing values associated with the second text strings with stored values; and
in response to determining one or more values that do not match, generating an error notification.
5. The method of claim 1 , wherein generating an output comprises:
inserting the second text strings into a second document;
inserting corresponding values for each of the second text strings and extracted from the received document into the second document; and
providing the second document to a sender of the received document.
6. 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 document having a plurality of first text strings each representing a respective description related to a real estate transaction and a control value associated with the real estate transaction and having a first value from which values associated with one or more descriptions of the plurality of first text strings depend;
extracting the plurality of first text strings and the control value from the document;
validating the values associated with the respective descriptions of the plurality of first text strings in the document, the validating comprising comparing the control value to a corresponding stored value;
in response to determining a match between the control value and the stored value, providing the extracted plurality of first text strings to a first machine learning model, wherein the first machine learning model is trained to output a numerical vector representation for each input first text string, wherein each vector representation represents a meaning of the corresponding description;
providing the output vector representations from the first machine learning model to a second machine learning model, wherein the second machine learning model is trained to output a second text string representing a respective title production system description related to the real estate transaction for each input vector representation such that for each description corresponding to an input first text string a corresponding title production system description is output; and
processing the second text strings to generate an output.
7. The system of claim 6 , wherein the first machine learning model is a document to vector model trained using a collection of text strings to generate vectors that represent the meaning of the text strings.
8. The system of claim 7 , wherein the second machine learning model is a classification model trained using a collection of vectors generated by the first machine learning model to generate a prediction of a particular type of a set of types.
9. The system of claim 6 , wherein generating an output comprises:
comparing values associated with the second text strings with stored values; and
in response to determining one or more values that do not match, generating an error notification.
10. The system of claim 6 , wherein generating an output comprises:
inserting the second text strings into a second document;
inserting corresponding values for each of the second text strings and extracted from the received document into the second document; and
providing the second document to a sender of the received document.
11. One or more non-transitory computer storage media encoded with computer program instructions that when executed by one or more computers cause the one or more computers to perform operations comprising:
receiving a document having a plurality of first text strings each representing a respective description related to a real estate transaction and a control value associated with the real estate transaction and having a first value from which values associated with one or more descriptions of the plurality of first text strings depend;
extracting the plurality of first text strings and the control value from the document;
validating the values associated with the respective descriptions of the plurality of first text strings in the document, the validating comprising comparing the control value to a corresponding stored value;
in response to determining a match between the control value and the stored value, providing the extracted plurality of first text strings to a first machine learning model, wherein the first machine learning model is trained to output a numerical vector representation for each input first text string, wherein each vector representation represents a meaning of the corresponding description;
providing the output vector representations from the first machine learning model to a second machine learning model, wherein the second machine learning model is trained to output a second text string representing a respective title production system description related to the real estate transaction for each input vector representation such that for each description corresponding to an input first text string a corresponding title production system description is output; and
processing the second text strings to generate an output.
12. The non-transitory computer storage media of claim 11 , wherein the first machine learning model is a document to vector model trained using a collection of text strings to generate vectors that represent the meaning of the text strings.
13. The non-transitory computer storage media of claim 12 , wherein the second machine learning model is a classification model trained using a collection of vectors generated by the first machine learning model to generate a prediction of a particular type of a set of types.
14. The non-transitory computer storage media of claim 11 , wherein generating an output comprises:
comparing values associated with the second text strings with stored values; and
in response to determining one or more values that do not match, generating an error notification.
15. The non-transitory computer storage media of claim 11 , wherein generating an output comprises:
inserting the second text strings into a second document;
inserting corresponding values for each of the second text strings and extracted from the received document into the second document; and
providing the second document to a sender of the received document.