We are building an automated Invoice processing application, that performs the following
- Extract text and its coordinates from Invoice documents. From this create a Pattern Structure (PS) or Keyword Structure (KWS) and map them into a graph that uniquely identifies the type of invoice.
- Utilize Case Base Reasoning (CBR) to build multiple such models for different invoice types. If a graph from a new invoice matches one of the models then the CBR process will know how to process it.
- Build a probabilistic classification model using previously processed invoice data to build a confidence score on the CBR processed invoice.
- If the score is below a defined threshold, flag it for a manual verification. Otherwise continue the rest of the workflow.
- Feed any processing errors back to the CBR model generator to increase processing accuracy with more data