Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

Combining Embeddings and Domain Knowledge for Job Posting Duplicate Detection

Matthias Engelbach, Dennis Klau, Maximilien Kintz, Alexander Ulrich

TL;DR

This work tackles the practical problem of detecting duplicate job postings across multiple sources by blending overlap-based text similarity, semantic embeddings, and domain-knowledge weighted keyword matching. The proposed multi-component pipeline, with data preprocessing and a final Total Score TS, significantly improves precision-recall balance and achieves an $F1$ of $0.94$ on validation, enabling productive deployment in a real-world recruitment setting. Key contributions include a robust integration of text overlap with skill-focused measures, a dynamic weighting scheme for skills, and a production-ready deployment that demonstrates scalability and real-user benefit. The results highlight the value of combining traditional text matching with domain-specific knowledge, while outlining avenues for future enhancements such as larger-scale evaluation and optional usage of large language models for scoring.

Abstract

Job descriptions are posted on many online channels, including company websites, job boards or social media platforms. These descriptions are usually published with varying text for the same job, due to the requirements of each platform or to target different audiences. However, for the purpose of automated recruitment and assistance of people working with these texts, it is helpful to aggregate job postings across platforms and thus detect duplicate descriptions that refer to the same job. In this work, we propose an approach for detecting duplicates in job descriptions. We show that combining overlap-based character similarity with text embedding and keyword matching methods lead to convincing results. In particular, we show that although no approach individually achieves satisfying performance, a combination of string comparison, deep textual embeddings, and the use of curated weighted lookup lists for specific skills leads to a significant boost in overall performance. A tool based on our approach is being used in production and feedback from real-life use confirms our evaluation.

Combining Embeddings and Domain Knowledge for Job Posting Duplicate Detection

TL;DR

This work tackles the practical problem of detecting duplicate job postings across multiple sources by blending overlap-based text similarity, semantic embeddings, and domain-knowledge weighted keyword matching. The proposed multi-component pipeline, with data preprocessing and a final Total Score TS, significantly improves precision-recall balance and achieves an of on validation, enabling productive deployment in a real-world recruitment setting. Key contributions include a robust integration of text overlap with skill-focused measures, a dynamic weighting scheme for skills, and a production-ready deployment that demonstrates scalability and real-user benefit. The results highlight the value of combining traditional text matching with domain-specific knowledge, while outlining avenues for future enhancements such as larger-scale evaluation and optional usage of large language models for scoring.

Abstract

Job descriptions are posted on many online channels, including company websites, job boards or social media platforms. These descriptions are usually published with varying text for the same job, due to the requirements of each platform or to target different audiences. However, for the purpose of automated recruitment and assistance of people working with these texts, it is helpful to aggregate job postings across platforms and thus detect duplicate descriptions that refer to the same job. In this work, we propose an approach for detecting duplicates in job descriptions. We show that combining overlap-based character similarity with text embedding and keyword matching methods lead to convincing results. In particular, we show that although no approach individually achieves satisfying performance, a combination of string comparison, deep textual embeddings, and the use of curated weighted lookup lists for specific skills leads to a significant boost in overall performance. A tool based on our approach is being used in production and feedback from real-life use confirms our evaluation.
Paper Structure (17 sections, 7 equations, 3 figures, 1 table)

This paper contains 17 sections, 7 equations, 3 figures, 1 table.

Table of Contents

  1. Introduction
  2. Industrial Use Case and Requirements
  3. Related Work
  4. Data Resources
  5. Method and Experimental Setting
  6. Data Preprocessing
  7. Evaluation Metrics
  8. Overlap-based Similarity Matching
  9. Embedding-based Similarity Matching
  10. Weighted Keyword Matching
  11. Combined Matching
  12. Results and Discussion
  13. Threshold Finding
  14. Experimental Results
  15. Productive Deployment
  16. ...and 2 more sections

Figures (3)

  • Figure 1: Example of job offering comparison for two different positions that are part of the the same project. While the original description text has greater parts of overlap in boiler plate text sections, the extracted pure skill texts differ significantly. Yellow markings indicate overlapping text parts between source and target texts. The censored parts mark confidential information.
  • Figure 2: Overview of the different scoring approaches in their respective matching classes and the final selected component of each class for the total score. The components were selected by their performance on the validation set.
  • Figure 3: Score predictions for the different methods on the test set with 74 duplicates and 73 non-duplicates. The thresholds are set manually for each score to gain a good distinction between positive and negative duplicate predictions.