Automated Information Extraction From Offer Letters Using Deep Learning: A Comparative Study

Read full text Summarise with VIRAT
Authors: Chandrashekhar Aazad, Ramneet Singh Chadha, Jasmehar Singh
Cited by: 0
References: 9

Summary

No summary yet. VIRAT can read this paper and write a plain-language summary of its aim, methods, key findings and limitations.

Summarise with VIRAT Free to try with your Science Catalogs account.

Abstract

The fast move to digital for HR tasks has led to the creation of a lot of unstructured documents, especially offer letters, which provide important information about employees. Using usual approaches, this has not been processed well for extracting information. This study presents a comprehensive system for the automatic extraction of information from offer letters utilizing optical character recognition (OCR) and named entity recognition. A comparison investigation has been conducted about the performance of the conventional Conditional Random Field (CRF) model and the advanced Bidirectional Encoder Representations from Transformers (BERT) model. This paper presents a system for generating synthetic data to address the challenges of data scarcity and privacy. This framework may create fake data for the process of getting information from offer letters, such as alternative layouts, semantic content, and noise. This paper evaluated the efficacy of the models on a demanding dataset of 500 documents. The findings indicate that, while the traditional Conditional Random Field model yielded satisfactory outcomes, the advanced Bidirectional Encoder Representations from Transformers model produced significantly superior results in comparison to the conventional model. This experimental evaluation also proves the efficiency of the proposed system in extracting key entities such as candidate name, designation, salary, and date of joining accurately from the scanned offer letters. The comparative analysis proves that, although the proposed system using the CRF model works accurately in structured data, the BERT model has a better capacity for understanding the contextual relationship within the document text. Thus, the proposed transformer model proves to be effective in dealing with the variability in the document layouts, linguistic expressions, and document conditions.

Fields of Study & Subjects

Handwritten Text Recognition Techniques Topic Modeling AI and HR Technologies

Publication details

Journal / source
International Journal For Multidisciplinary Research
Volume
8
Issue
2
Publication date
21 March 2026
Language
EN