ViReCAX: Vietnamese - Recruiment - Abnormal - Explanation dataset

This repository contains code for our paper: Vuong-Quoc Nguyen Huynh, Bao-Khanh Truong Phuoc, Anh Hoang Tran, Duong Tien Pham, Son T. Luu, Trong-Hop Do. Abnormal Job Postings Detection with Explanation on Vietnamese Job Hunter Platforms was published at Neural Computing and Applications journal.

The repository is structured as follows:

• baseline: Contains experiments on offline learning baseline models on the dataset.
• system: Contains experiments on online learning systems with three different strategies.

To access the dataset, please download the ViReCAX_User_Agreement.docx from this repository, fill and sign it, and send the signed document back in PDF format to the following email addresses:

• Mr. Luu Thanh Son at [email protected]
• Mr. Nguyen Huynh Vuong Quoc at [email protected]

Abstract

Recruitment is a crucial part of human resources for businesses and organizations. The growth of online recruitment websites has made them a popular way to find potential candidates. However, this increase has also led to a rise in fraudulent job postings, which can result in losses of personal information, assets, and company reputation. As a result, with advancements in AI, there is a need for tools to evaluate the verification of job postings. Hence, in this paper, we introduce the ViReCAX (Vietnamese Recruiment - Abnormal - Explanation) dataset, a novel dataset designed for detecting abnormal job postings on Vietnamese job hunter platforms, with an emphasis on providing explanations for the model's decisions. The dataset comprises 12,054 manually annotated job postings, covering three tasks: job posting classification (CLEAN, WARNING, SEEDING), aspect-level verification (POSITIVE, NEGATIVE, NEUTRAL, NOT-MENTIONED), and explanation generation. Inter-annotator agreement scores were moderate for the classification tasks (0.6 and 0.59) and high for the text generation task (BLEU-2 score of 0.74 and BERTScore of 0.73). We also propose baseline models for detecting abnormal job postings, utilizing BERT-based models with LSTM or CNN for classification tasks and sequence-to-sequence models for the generation task. Finally, we propose three different online streaming learning strategies to improve the model's ability to adapt to streaming data scenarios.

Data Sample

• 10 data samples from train dataset: ./data-sample/train_sample_10.csv

Citation

For any usage related to all codes and data used from our repository, please cite our following paper:

@article{Nguyen2026AbnormalJobPosting,
  author = {Vuong-Quoc Huynh Nguyen and Bao-Khanh Phuoc Truong and Anh Hoang Tran and Duong Tien Pham and Son T. Luu and Trong-Hop Do},
  title = {Abnormal Job Postings Detection with Explanation on Vietnamese Job Hunter Platforms},
  journal = {Neural Computing and Applications},
  year = {2026},
  publisher = {Springer},
  doi = {10.1007/s00521-026-12153-5},
  url = {https://link.springer.com/article/10.1007/s00521-026-12153-5}
}

For any questions, please contact our corresponding author:

• Mr. Luu Thanh Son at [email protected]
• Mr. Nguyen Huynh Vuong Quoc at [email protected] for any question about offline learning baseline models and dataset
• Mr. Tran Hoang Anh at [email protected] for any question about online learning systems