[Paper Review] Learning Effective Representations for Person-Job Fit by Feature Fusion

Junshu Jiang, Songyun Ye+3 more|arXiv (Cornell University)|Jun 12, 2020

Topic Modeling38 references4 citations

TL;DR

This paper proposes PJFFF, a feature fusion method for person-job fit that combines explicit representations from semantic entities (e.g., education, skills) and deep text features via CNN and DeepFM, with implicit representations from historical application data using LSTM. The model achieves state-of-the-art performance on real-world recruitment data, outperforming existing methods with significant margins and enabling interpretable matching through entity-level analysis.

ABSTRACT

Person-job fit is to match candidates and job posts on online recruitment platforms using machine learning algorithms. The effectiveness of matching algorithms heavily depends on the learned representations for the candidates and job posts. In this paper, we propose to learn comprehensive and effective representations of the candidates and job posts via feature fusion. First, in addition to applying deep learning models for processing the free text in resumes and job posts, which is adopted by existing methods, we extract semantic entities from the whole resume (and job post) and then learn features for them. By fusing the features from the free text and the entities, we get a comprehensive representation for the information explicitly stated in the resume and job post. Second, however, some information of a candidate or a job may not be explicitly captured in the resume or job post. Nonetheless, the historical applications including accepted and rejected cases can reveal some implicit intentions of the candidates or recruiters. Therefore, we propose to learn the representations of implicit intentions by processing the historical applications using LSTM. Last, by fusing the representations for the explicit and implicit intentions, we get a more comprehensive and effective representation for person-job fit. Experiments over 10 months real data show that our solution outperforms existing methods with a large margin. Ablation studies confirm the contribution of each component of the fused representation. The extracted semantic entities help interpret the matching results during the case study.

Motivation & Objective

• To address the limitations of existing person-job fit methods that rely solely on free-text processing and ignore structured resume/job post fields.
• To improve matching accuracy by capturing both explicit information (e.g., skills, degrees) and implicit intentions (e.g., recruiter preferences) from historical application data.
• To enhance model interpretability by extracting and utilizing semantic entities that can explain matching decisions.
• To develop an efficient, scalable solution suitable for real-world deployment on large-scale recruitment platforms.

Proposed method

• Extract semantic entities (e.g., university, work experience, skills) from full resumes and job posts using NLP techniques, then embed them into a vector for explicit feature learning.
• Apply a convolutional neural network (CNN) to learn deep representations from free-text fields (e.g., work experience, job description).
• Use a DeepFM model to learn cross-feature interactions among the extracted semantic entities, capturing high-order relationships.
• Process historical application sequences (accepted/rejected resumes) using a unidirectional LSTM to model candidate and recruiter implicit intentions.
• Fuse explicit features (from entities and text) and implicit features (from LSTM) via concatenation to form a comprehensive representation for matching.
• Optimize the final model for online deployment using CPU-based inference, Go-based pre-processing, and knowledge distillation to reduce latency.

Experimental results

Research questions

• RQ1 Can combining structured semantic entities with deep text representations improve person-job fit matching performance?
• RQ2 To what extent do historical application patterns reveal implicit recruiter or candidate preferences not stated in job posts or resumes?
• RQ3 How effective is the fusion of explicit and implicit representations in enhancing matching accuracy and interpretability?
• RQ4 Can the proposed method achieve high performance while remaining efficient and scalable for real-world deployment?

Key findings

• The proposed PJFFF model outperforms existing methods by a large margin on 10 months of real-world recruitment data, demonstrating superior matching accuracy.
• Ablation studies confirm that each component—semantic entities, text features, and implicit intention modeling—contributes positively to the final performance.
• The inclusion of semantic entities significantly improves model performance, as shown by the comparison between models using only text and those using both text and entities.
• The LSTM-based modeling of historical applications captures implicit intentions, leading to measurable performance gains over models that ignore such data.
• The model’s interpretability is enhanced by the extracted semantic entities, which clearly explain why certain candidates are ranked higher or lower.
• The final model achieves inference times of 0.2–0.3 seconds per sample after optimization, making it suitable for large-scale online deployment.