Text Annotation and OCR-Based Named Entity Recognition - India, UK and USA

Client Background

The client needed to have high quality structured datasets based on unstructured document images that would be used downstream, including data analytics and AI model training, and automated information extraction. The documents had very vital contents like names, locations, organizations, dates, and financial contents which had to be well identified and classified.

Business Challenge

The client had a number of operational and technical difficulties:

• Handling unstructured text in large quantities of image-based documents.
• Verifying a high level of accuracy in OCR extraction of mixed-image quality and forms.
• Named entities are consistently identified and labelled in documents.
• Transforming free-form text to machine-readable Excel formats.
• Data consistency and the reduction of manual processing errors.

The client needed a scalable partner, which would integrate OCR processing with text annotation domain knowledge.

Infosearch Solution

Infosearch BPO provided an end-to-end solution in relation to processing of OCR, structuring of data, and annotation of texts. The range of services covered was:

• Moving graphic files into searchable PDF files through the use of OCR.
• Extraction of the text content and accuracy checking of the OCR output.
• Conducting Named Entity Recognition (NER) to recognize entities, like persons, organizations, locations, dates, and important data.
• Converting extracted entities into analyzable Excel formats to be used in model training.
• The introduction of quality checks to ascertain data precision and similarity.

Infosearch created trained data annotation experts and used standard workflows to provide reliable and scalable delivery.

Approach and Methodology

Infosearch adhered to a developed workflow in order to achieve high-quality results:

• Document Assessment - Measuring quality of images, form and organizations of a document.
• OCR Processing -Image to searchable PDFs through state-of-the-art OCR software.
• Data Validation -Manual and automated tests to eliminate OCR errors.
• Named Entity Annotation– This is the process of identifying and classifying objects according to a set of existing taxonomy.
• Data Structuring- However, annotated data was converted into structured Excel formats.
• Quality Assurance- Benchmarking and multi-level reviews.

Outcomes and Results

The work produced quantifiable value to the client:

• Text extraction and high-precision OCR conversion.
• Named entity recognition in a consistent and reliable fashion.
• Automated datasets to analytics and AI.
• Less manual processing of data.
• Increased volume in the processing of large documents.

Business Impact

Outsourcing the services of OCR and text annotation to Infosearch BPO allowed the client to enhance the efficiency of its operations and speed up the availability of the data to be used in machine learning and analytics projects. The organized Excel results allowed the user to integrate the existing systems easily and to make better decisions.

Conclusion

The case study illustrates that Infosearch BPO has mastered the use of OCR technology coupled with text annotation and Named Entity Recognition. The interaction also demonstrates that the organization can convert unstructured image-based data into structured high-quality data that can be used in advanced analytics and AI-based applications.