Overview
Introduction
Service Features
Service Process
Output Quality
 |
 |
|
|
OUTPUT QUALITY Beginner Automated processing must rely heavily on OCR. Does poor quality OCR lower your accuracy? For paper documents, the auto-encoding engine does rely on OCR output. Obviously, higher OCR quality allows for better coverage and accuracy. However, even with OCR output of only moderate quality, the software is able to extract accurate information from the OCR text. This is possible because the system combines a sophisticated OCR-correction utility that compensates for OCR mistakes with artificial intelligence and statistical analysis techniques. For example, the software can identify the phrase "FACSJMILE" as "facsimile" and encode the document type accordingly. Similarly, it can tell that "MarshaU" in the OCR text was "Marshall" in the original document and that "Robed" should be encoded as "Robert." Such correction techniques boost the accuracy of the output significantly, and result in an error rate that is usually below 3%. (For reference, our experience has shown that typical human coding error rates range between 7-10%.) Intermediate How does the quality of automated encoding compare to manual encoding? Automated encoding uses a combination of computer processing and human judgment. In the early parts of the process, a trained operator assists software to produce high quality first-pass output. A Quality Control team then reviews each record of this data using computer-assisted tools to accelerate the process. The result is a database of comparable or better quality to one produced by hand. Manual encoding frequently suffers from consistency issues, because many coders work on the same job. The automated solution uses many fewer people to process the same amount of data, and computer-aided tools help ensure consistency both within the case and across cases. How can you ensure 98% accuracy? How do you measure accuracy? Valora's output is produced by a proprietary, multi-pass process that allows us to encode many documents, very accurately, with only a few people. This process includes a number of computer-assisted tools designed to ensure accuracy and consistency. After processing, however, the data undergoes an internal audit that provides an additional check to ensure that the process has completed successfully and the data meets the project specification and accuracy requirements. Any data volume that does not pass the audit is reprocessed. Valora uses the strictest accuracy measure possible. That is, that every character of every field of every record is either perfectly correct, or the entire record is considered inaccurate. For 98% accuracy, 98 out of every 100 records will be absolutely perfect and the remaining 2 records will each have only one error (one incorrect character). Advanced If Valora checks every record for accuracy, isn't the Quality Control process a bottleneck? Because of Valora's sophisticated tools, our Quality Control engineers can validate documents at many times the rate that manual coders can enter data in the first place. Valora's process typically yields between a 4:1 and 5:1 efficiency increase, meaning we accomplish with 1 person, what others require 4 or 5 people to do. The result is that high-quality data can be delivered on a schedule that fits with most customer requirements. |
||
 |
 |