Automated financial data aggregation

1. A computer-implemented method, comprising:

crawling a financial institution website to identify a plurality of web pages in the financial institution website;

for each webpage of at least some of the plurality of web pages, determining a category for the webpage from a plurality of web page categories by inputting the web page into a web page classifier that includes a first data model that has been trained using a machine learning technique based on a plurality of webpages that are not from the financial institution website, the plurality of web page categories including at least a login page, an accounts summary page, and an account page;

for each segment of at least some segments from the plurality of web pages, determining a category for the segment from a plurality of segment categories by inputting the segment into a segment classifier that includes a second data model that has been trained using a machine learning technique based on a plurality of segments that are not from the financial institution website, the plurality of segment categories including at least a login field, an accounts description table, and a transaction descriptions table;

generating, automatically and in the computer, a site-specific script for extracting financial data from the financial institution website based on the determined categories for the plurality of web pages and segments in the financial institution website; and

extracting, in the computer, for one or more users, financial data from the financial institution website using the generated site-specific script.

2. The method of claim 1 , wherein the first data model implements a naive Bayes classification.

3. The method of claim 1 , wherein determining the category for the webpage includes:

determining a respective plurality of scores for the web page, each score in the plurality of scores indicating a confidence that the web page corresponds to a particular category;

determining for each web page whether a score in the respective plurality of scores satisfies a threshold; and

in response to determining that a score in the respective plurality of scores for the web page satisfies a threshold, associating the web page with a particular category corresponding to the score.

4. The method of claim 3 , further comprising:

in response to determining that no score in the respective plurality of scores for a web page satisfies a threshold, providing the web page to a user for manual categorization; and

associating the web page with a category specified by the user.

5. The method of claim 1 , wherein the second data model implements a nave Bayes classification.

6. The method of claim 1 , wherein determining the category for the segment includes:

determining a respective plurality of scores for the segment, each score in the plurality of scores indicating a confidence that the segment corresponds to a particular category;

determining for each segment whether a score in the respective plurality of scores satisfies a threshold; and

in response to determining that a score in the respective plurality of scores for the segment satisfies a threshold, associating the segment with the a particular category corresponding to the score.

7. The method of claim 6 , further comprising:

in response to determining that no score in the respective plurality of scores for a segment satisfies a threshold, providing the segment to a user for manual categorization; and

associating the segment with a category specified by the user.

8. The method of claim 1 , further comprising:

prior to crawling the financial institution website determining, in the computer, that a site-specific script for extracting financial data from the financial institution website is not available.

9. A computer program product, comprising a non-transitory computer-readable medium encoded with instructions to cause a data processing apparatus to:

crawl a financial institution website to identify a plurality of web pages in the financial institution website;

for each webpage of at least some of the plurality of web pages, determine a category for the webpage from a plurality of web page categories by inputting the web page into a web page classifier that includes a first data model that has been trained using a machine learning technique based on a plurality of webpages that are not from the financial institution website, the plurality of web page categories including at least a login page, an accounts summary page, and an account page;

for each segment of at least some segments from the plurality of web pages, determine a category for the segment from a plurality of segment categories by inputting the segment into a segment classifier that includes a second data model that has been trained using a machine learning technique based on a plurality of segments that are not from the financial institution website, the plurality of segment categories including at least a login field, an accounts description table, and a transaction descriptions table;

automatically generate a site-specific script for extracting financial data from the financial institution website based on the categories for the plurality of web pages and segments in the financial institution website; and

extracting, for one or more users, financial data from the financial institution website using the generated site-specific script.

10. The computer program product of claim 9 , wherein the first data model implements a naive Bayes classification.

11. The computer program product of claim 9 , wherein the instructions to determine the category for the webpage include instructions to:

determine a respective plurality of scores for the web page, each score in the plurality of scores indicating a confidence that the web page corresponds to a particular category;

determine for each web page whether a score in the respective plurality of scores satisfies a threshold; and

in response to determining that a score in the respective plurality of scores for the web page satisfies a threshold, associate the web page with a particular category corresponding to the score.

12. The computer program product of claim 11 , further comprising instructions to:

in response to determining that no score in the respective plurality of scores for a web page satisfies a threshold, provide the web page to a user for manual categorization; and

associate the web page with a category specified by the user.

13. The computer program product of claim 9 , wherein the second data model implements a naive Bayes classification.

14. The computer program product of claim 9 , wherein the instructions to determine the category for the segment include instructions to:

determine a respective plurality of scores for the segment, each score in the plurality of scores indicating a confidence that the segment corresponds to a particular category;

determine for each segment whether a score in the respective plurality of scores satisfies a threshold; and

in response to determining that a score in the respective plurality of scores for the segment satisfies a threshold, associate the segment with the a particular category corresponding to the score.

15. The computer program product of claim 14 , further comprising instructions to:

in response to determining that no score in the respective plurality of scores for a segment satisfies a threshold, provide the segment to a user for manual categorization; and

associate the segment with a category specified by the user.

16. The computer program product of claim 9 , further comprising instructions to:

prior to crawling the financial institution website, determine that a site-specific script for extracting financial data from the financial institution website is not available.