Scientific Document Extraction

Helped a leading pharmaceutical company to digitize its “PlaceHolder Document Creation” to reduce processing time and manpower costs

Scientific-Document-Extraction-banner

About the Client

The client is an Indian multinational pharmaceutical company. The client had recently acquired another leading pharmaceutical company and was going through a major phase of digitization of its existing documentation and related processes.

The Business Challenge

The client had thousands of medicine preparation templates. The client was updating the values in the blank spaces manually that was taking a long time, and prone to human error.
As part of the Quality Systems initiative (a program where all existing documents are digitized and end-users are given an interface), the client wished to automate the following process:

  • Replacement of the placeholders (Blank Spaces) in the template files (ODT format) with some meaningful texts (chemical entity or process) in between even numbers of #s
  • Process execution for individual documents and folders containing multiple documents

What Aptus Data Labs Did

We developed a traditional Named Entity Recognition model using POS tags and another using LSTM. Both models were able to detect the meaningful chemical entities on either side of the blank spaces in the document and replaced the blanks with these texts.

The Business Impact Aptus Data Labs Made

With the entire manual process successfully automated, the manual effort was reduced to a staggering 5% of the initial, and human resources costs were reduced greatly as well.

The solution was deployed on Rapidminer to provide a seamless interface to run the process for 100s of documents. As a result, the time for the entire process was reduced from hours to minutes.

The Business and Technology Approach

Aptus Data Labs used the following methodology for automating the process to resolve the existing challenge. Aptus Data Labs:

  • Looked for meaningful text in both the left & right side of blank spaces to fill all spaces with meaningful/relevant chemical names or phrases
  • Tried to narrow down the search space to avoid unnecessary or wrong information as placeholder text
  • Detected a small window of neighboring texts to provide information about the placeholder text
  • Used the POS Tag and LSTM-based Named Entity Recognition models to extract accurate placeholder text from the window
  • Replaced blank spaces with extracted text to not lose the semantic sense of the sentence
  • Repeated the above process for all the spaces
  • Operationalized the entire text analytics model in Rapidminer

Tools used

Python, Gensim, Keras, Powershell Scripts and Rapidminer.

The Outcome

  • Automating the manual process using Machine Learning helped the client save costs on human resources.
  • The deliverables accuracy increased by 90% to 95%.
  • Helped the client to reduce the time to complete the processing of batches of documents.

Related Case Studies

Download Case study

Explore more of our client stories

Download Case study