# Construction of English Resume Corpus and Test with Pre-trained Language Models

**Authors:** Chengguang Gan, Tatsunori Mori  
**arXiv:** [2208.03219](https://arxiv.org/abs/2208.03219)  
**Date:** 2022-08-05  
**Summary page:** https://www.summarizepaper.com/en/paper/2208.03219v2/

## Key points

- Natural Language Processing (NLP) relies on Information Extraction (IE)
- IE is critical for extracting information from resumes
- Constructed text can be obtained and stored for later search and analysis
- AI resume screening systems can reduce HR labor costs
- The study by Chengguang Gan and Tatsunori Mori transforms the task of information extraction into a simple sentence classification task
- The authors improve upon the English resume dataset produced by prior studies by creating larger and more fine-grained classification rules, resulting in an improved accuracy rate for the original dataset
- The refined dataset is used to test current mainstream Pre-training language models (PLMs)
- Experiments are conducted to explore the relationship between training sample size and correctness rate for the dataset
- This study provides valuable insights into improving information extraction from resumes using advanced NLP techniques that could have significant implications for HR departments worldwide.
- These techniques could help reduce time spent manually reviewing resumes while ensuring accuracy in identifying qualified candidates.

## Summary

The field of Natural Language Processing (NLP) has always relied on Information Extraction (IE) as one of its essential tasks. One of the most critical application scenarios for IE is the extraction of information from resumes. By classifying each part of a resume, constructed text can be obtained and stored for later search and analysis. This data can also be used in AI resume screening systems to significantly reduce HR labor costs. In this study by Chengguang Gan and Tatsunori Mori, the task of information extraction from resumes is transformed into a simple sentence classification task. The authors improve upon the English resume dataset produced by prior studies by creating larger and more fine-grained classification rules, resulting in an improved accuracy rate for the original dataset. The refined dataset is then used to test current mainstream Pre-training language models (PLMs). Additionally, experiments are conducted to explore the relationship between training sample size and correctness rate for the dataset. Overall, this study provides valuable insights into improving information extraction from resumes using advanced NLP techniques that could have significant implications for HR departments worldwide. These techniques could help reduce time spent manually reviewing resumes while ensuring accuracy in identifying qualified candidates.