Research on Model Optimization Using Deep Learning in Intelligent Employment Guidance Systems

View at publisher Summarise with VIRAT
Authors: Wei Zhou, Yalan Cheng
Cited by: 0

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

This paper proposes a dynamic attention fusion employment guidance model (DAF-EGM). This model optimizes matching performance through the collaboration of multiple modules: a weighted fusion CNN with multiple convolutional kernels extracts implicit skills and collaboration requirements from job descriptions. An LSTM with a temporal decay mechanism captures dynamic trends in job seekers' career histories. A dynamic attention module is constructed by combining a time decay factor (to adapt to changes in job search demand) with a job competitiveness coefficient (to quantify market dynamics such as the number of applicants and supply-demand ratio) to achieve adaptive adjustment of feature weights. Experiments are conducted on 72,863 valid samples preprocessed from three types of real-world data: 50,000 recruitment platform data, 10,000 graduate tracking data, and supply-demand data from 12 industries. The model is compared with traditional collaborative filtering (CF), static attention deep learning (SA-DL), and CNN-RNN fusion (CR-Fusion) models. Results show that in the Top-10 recommendation scenario, DAF-EGM achieves 89.2% precision, 86.7% recall, 87.8% F1-score, and 0.083 MAE. These improvements are 26.9, 28.2, and 27.6 percentage points, respectively, compared to traditional CF, with a MAE reduction of 0.132. Compared to SA-DL, DAF-EGM achieves a 13.6 percentage point improvement in precision and a MAE reduction of 0.075. Simulation plots demonstrate that DAF-EGM's precision decreases by only 3.2% with increasing Top-K (5-25) (compared to 7.4% for traditional CF), and MAE drops below 0.1 after 15 training epochs (compared to 25 for SA-DL). Ablation experiments confirm that the time decay factor and the job competition coefficient synergistically improve F1-score by over 14 percentage points, providing technical support for the intelligent upgrade of employment guidance systems.

Fields of Study & Subjects

Recommender Systems and Techniques AI and HR Technologies Information Retrieval and Search Behavior

Publication details

Journal / source
IEEE International Conference on Power, Electronics and Green Energy (ICPEGE)
Pages
706–710
Publication date
28 January 2026
Language
EN