Systems and methods for training document analysis system for automatically extracting data from documents

Abstract

A method of training a document analysis system to extract data from documents is provided. The method includes: automatically analyzing images and text features extracted from a document to associate the document with a corresponding document category; comparing the extracted text features with a set of text features associated with corresponding category of the document, in which the set of text features includes a set of characters, words, and phrases; if the extracted features are found to consist of the characters, words, and phrases belonging to the set of text features associated with the corresponding document category, storing the extracted text features as the data contained in the corresponding document; and, if the extracted text features are found to include at least one text feature that does not belong to the set of text features associated with the corresponding document category, submitting the unrecognized text features to a training phase.

Description

CROSS REFERENCE TO RELATED APPLICATIONS [0001]This application claims the benefit under 35 U.S.C. §119(e) of U.S. Provisional Patent Application No. 61 / 295,210, filed Jan. 15, 2010, which is hereby incorporated by reference herein in its entirety.[0002]This application is also related to the following applications filed concurrently herewith on Jan. 14, 2011:[0003]U.S. patent application Ser. No. ______, entitled “Systems and methods for automatically extracting data from electronic documents containing multiple layout features;”[0004]U.S. patent application Ser. No. ______, entitled “Systems and methods for automatically extracting data from electronic documents using external data;”[0005]U.S. patent application Ser. No. ______, entitled “Systems and methods for automatically correcting data extracted from electronic documents using known constraints for semantics of extracted data elements;”[0006]U.S. patent application Ser. No. ______, entitled “Systems and methods for automatica...

AI Technical Summary

Benefits of technology

[0039]If the extracted text features are found to consist of the characters, words, and phrases belonging to the set of text features associated with the corresponding electronic document category, the method further includes storing the extracted text features as the data contained in the corresponding electronic document. If, however, the extracted text features are found to include at least one text feature that does not belong to the set of text features associated with the corresponding electronic document category, the method further includes submitting the unrecognized text features to a training phase in which the text features are recognized as belonging to the set of text features associated with the corresponding electronic document category and then using the now-recognized text features to automatically modify the set of text features associated with the corresponding electronic document category so that the extracting data, regardless of which document category the corresponding document belongs to, improves as the training method is subjected to more and more unrecognized text features and the set of text features is modified accordingly.

Problems solved by technology

Electronic Data Interchange is known for custom computer systems, cumbersome software and bloated standards that defeated its rapid spread throughout the supply chain.

Perceived as too expensive, the vast majority of businesses have avoided implementing EDI.

Similarly, applications of XML, XBRL and other computer-readable document files are quite limited compared to the use of documents in paper and digital image formats (such as PDF and TIFF.)

Such manual data extraction is complex, time-consuming and error-prone.

As a result, the cost of data extraction is often quite high; numerous studies estimate the cost of processing invoices in excess of ten dollars each.

The cost is especially high when the data extraction is performed by accountants, lawyers, physicians and other highly paid professionals as part of their work.

Despite the potential productivity gains that are enabled with workflow software in the form improved labor utilization, manual document processing remains a fundamentally expensive process.

Since outsourcing is manual, just as is conventional data extraction, it is also complex, time-consuming and error-prone.

Quality problems with offshore data extraction work have been reported by many customers.

These measures reduce the cost savings expected from offshore outsourcing.

Outsourcing and offshoring are accompanied with concerns over security risks associated with fraud and identity theft.

Although the transmission of scanned image files to the data extraction organization may be secured by cryptographic techniques, the sensitive data and personal identifying information are in the clear, i.e., unencrypted, when read by data extraction workers prior to entry in the appropriate computer systems.

Many data extraction organizations claim to strictly limit physical access to the rooms in which the employees enter the data; further, such rooms may be isolated.

Since such seemingly comprehensive security precautions are primarily physical in nature, they are imperfect.

Because of these imperfections, lapses in physical security have occurred.

The owners, managers, staff, guards and contractors of data extraction organizations may misuse some or all of the unencrypted confidential information in their care.

Further, breaches of physical and information system security by external parties can occur.

Because data extraction organizations are increasingly located in foreign countries, there is often little or no recourse for American citizens victimized in this manner.

Because such customization projects often cost upwards of hundreds of thousands of dollars, data extraction automation is usually limited to large organizations that can afford significant capital investments.

Optical character recognition is imperfect, often mistaking more than one percent of the characters on clean, high quality documents.

Many documents are neither clean nor high quality, suffering from being folded or marred before scanning, distorted during scanning and degraded during post-scanning binarization.

As a result, some of the labels needed to identify data are often not recognizable; therefore, some of the data cannot be automatically extracted.

When a wide range of forms exists, such as the 10,000 plus variations of W-2, 1099, K-1 and other personal income tax forms, automated data extraction is quite limited.

Despite years of efforts, several tax document automation vendors claim 50% or less data extraction and admit to numerous errors with conventional data extraction methods.

Because automation requires human inspection, source documents with sensitive information are exposed in their entirety to data extraction workers.

Images