Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

JobFormer: Skill-Aware Job Recommendation with Semantic-Enhanced Transformer

Zhihao Guan, Jia-Qi Yang, Yang Yang, Hengshu Zhu, Wenjie Li, Hui Xiong

TL;DR

JobFormer tackles the challenge of skill-aware job recommendation under sparse JD content and heterogeneity between JD and user profiles. It introduces a semantic-enhanced transformer with local-global attention and a two-stage learning strategy that uses user skill distributions to guide JD recall and user profiles to drive CTR-based ranking. The work contributes a novel item-level encoder, a local-global attention mechanism, and joint losses for skill correlation and relation consistency, with extensive experiments on real-world and public datasets showing improved recall and ranking and enhanced interpretability. The approach offers a practical privacy-conscious framework for embedding rich semantics in JDs and delivering personalized job recommendations.

Abstract

Job recommendation aims to provide potential talents with suitable job descriptions (JDs) consistent with their career trajectory, which plays an essential role in proactive talent recruitment. In real-world management scenarios, the available JD-user records always consist of JDs, user profiles, and click data, in which the user profiles are typically summarized as the user's skill distribution for privacy reasons. Although existing sophisticated recommendation methods can be directly employed, effective recommendation still has challenges considering the information deficit of JD itself and the natural heterogeneous gap between JD and user profile. To address these challenges, we proposed a novel skill-aware recommendation model based on the designed semantic-enhanced transformer to parse JDs and complete personalized job recommendation. Specifically, we first model the relative items of each JD and then adopt an encoder with the local-global attention mechanism to better mine the intra-job and inter-job dependencies from JD tuples. Moreover, we adopt a two-stage learning strategy for skill-aware recommendation, in which we utilize the skill distribution to guide JD representation learning in the recall stage, and then combine the user profiles for final prediction in the ranking stage. Consequently, we can embed rich contextual semantic representations for learning JDs, while skill-aware recommendation provides effective JD-user joint representation for click-through rate (CTR) prediction. To validate the superior performance of our method for job recommendation, we present a thorough empirical analysis of large-scale real-world and public datasets to demonstrate its effectiveness and interpretability.

JobFormer: Skill-Aware Job Recommendation with Semantic-Enhanced Transformer

TL;DR

JobFormer tackles the challenge of skill-aware job recommendation under sparse JD content and heterogeneity between JD and user profiles. It introduces a semantic-enhanced transformer with local-global attention and a two-stage learning strategy that uses user skill distributions to guide JD recall and user profiles to drive CTR-based ranking. The work contributes a novel item-level encoder, a local-global attention mechanism, and joint losses for skill correlation and relation consistency, with extensive experiments on real-world and public datasets showing improved recall and ranking and enhanced interpretability. The approach offers a practical privacy-conscious framework for embedding rich semantics in JDs and delivering personalized job recommendations.

Abstract

Job recommendation aims to provide potential talents with suitable job descriptions (JDs) consistent with their career trajectory, which plays an essential role in proactive talent recruitment. In real-world management scenarios, the available JD-user records always consist of JDs, user profiles, and click data, in which the user profiles are typically summarized as the user's skill distribution for privacy reasons. Although existing sophisticated recommendation methods can be directly employed, effective recommendation still has challenges considering the information deficit of JD itself and the natural heterogeneous gap between JD and user profile. To address these challenges, we proposed a novel skill-aware recommendation model based on the designed semantic-enhanced transformer to parse JDs and complete personalized job recommendation. Specifically, we first model the relative items of each JD and then adopt an encoder with the local-global attention mechanism to better mine the intra-job and inter-job dependencies from JD tuples. Moreover, we adopt a two-stage learning strategy for skill-aware recommendation, in which we utilize the skill distribution to guide JD representation learning in the recall stage, and then combine the user profiles for final prediction in the ranking stage. Consequently, we can embed rich contextual semantic representations for learning JDs, while skill-aware recommendation provides effective JD-user joint representation for click-through rate (CTR) prediction. To validate the superior performance of our method for job recommendation, we present a thorough empirical analysis of large-scale real-world and public datasets to demonstrate its effectiveness and interpretability.
Paper Structure (22 sections, 12 equations, 6 figures, 4 tables)

This paper contains 22 sections, 12 equations, 6 figures, 4 tables.

Table of Contents

  1. Introduction
  2. Related Work
  3. Job Recommendation Systems
  4. JD Representation with Deep Learning
  5. Preliminaries
  6. JobFormer
  7. Item-Level Encoder
  8. Semantic-Enhanced Transformer
  9. Skill-Aware Representation Learning for JD Recall
  10. Click-Through Rate Prediction for JD Ranking
  11. Experiments
  12. Data Description
  13. Implementations
  14. Baseline Approaches
  15. JD Recall Performance
  16. ...and 7 more sections

Figures (6)

  • Figure 1: A motivating example of two-stage skill-aware job recommendation (a). The JD contains multiple items including duties and requirements (b), and the corresponding user can be represented with personal skill distribution (c). Note that actually the items in JD correspond to the skill labels, and the degrees of demands correspond to the skill distributions.
  • Figure 2: An illustration of the proposed JobFormer, which includes a recall stage for candidate JDs generation and a ranking stage for candidate JDs ranking. In the recall stage, the JD and its neighbors constitute the JD tuple, and the item-level encoder aims for the item representation, which acts as the input token for the semantic-enhanced transformer. Then the designed semantic-enhanced transformer encodes both the intra-job and inter-job information to acquire more discriminative JD representation, which is further recalled as candidate JDs according to the JD-user cosine similarity. Lastly, in the ranking stage, recalled candidate JDs are combined with user profiles for CTR prediction via a click predictor and ranked for a personalized job recommendation.
  • Figure 3: (Best viewed in color when zoomed in.) Qualitative success results of JD recall given user queries. For each user query, we show the top-3 ranked JD text. The first two rows exhibit the results of JobFormer, and the last two rows give the results of the state-of-the-art NRMS model. We observe that our JobFormer can find the correct results (i.e., red marked) in the first-ranked JDs, and NRMS is inferior to JobFormer.
  • Figure 4: Influence of Balance Parameters. The figures in the first row are the results of $\lambda$, and the second row gives the results of $\mu$.
  • Figure 5: (Best viewed in color when zoomed in.) The example of interpretable CTR prediction.
  • ...and 1 more figures