Professional Network Matters: Connections Empower Person-Job Fit

TL;DR

This paper tackles the underutilization of professional networks in Person-Job Fit by proposing a two-stage framework that combines a Workplace Heterogeneous Information Network (WHIN) with a Contextual Social Attention Graph Neural Network (CSAGNN). WHIN provides rich, heterogeneous representations through metapath-based pre-training, while CSAGNN applies a job-specific attention mechanism to filter noise from professional connections and infuse context from connected members. The approach yields superior results over strong baselines on three real LinkedIn datasets, with ablations confirming the value of WHIN pre-training and the job-specific attention. The work highlights practical impact for recruitment platforms by improving match quality and offering avenues for skill completion and connection-aware recommendations, while acknowledging privacy and fairness considerations.

Abstract

Online recruitment platforms typically employ Person-Job Fit models in the core service that automatically match suitable job seekers with appropriate job positions. While existing works leverage historical or contextual information, they often disregard a crucial aspect: job seekers' social relationships in professional networks. This paper emphasizes the importance of incorporating professional networks into the Person-Job Fit model. Our innovative approach consists of two stages: (1) defining a Workplace Heterogeneous Information Network (WHIN) to capture heterogeneous knowledge, including professional connections and pre-training representations of various entities using a heterogeneous graph neural network; (2) designing a Contextual Social Attention Graph Neural Network (CSAGNN) that supplements users' missing information with professional connections' contextual information. We introduce a job-specific attention mechanism in CSAGNN to handle noisy professional networks, leveraging pre-trained entity representations from WHIN. We demonstrate the effectiveness of our approach through experimental evaluations conducted across three real-world recruitment datasets from LinkedIn, showing superior performance compared to baseline models.

Figures (6)

• Figure 1: The metagraph of Workplace Heterogeneous Information Network. It encompasses not only members (M) and jobs (J), which are crucial for Person-Job Fit, but also entities such as skills (S), companies (C), and schools (H).
• Figure 2: Steps of WHIN pre-training. (a) Workplace heterogeneous graph with metapath. (b) Subgraph sampling for mini-batch pre-training. (c) A pre-training model with encoder-decoder architecture using Link-level pre-training task.
• Figure 3: Architecture of CSAGNN. When determining the match between job $j_0$ and member $m_0$, the information from $m_0$'s professional connections, $m_1$ and $m_2$, is simultaneously acquired. The initial representations of both the member and job are formed by concatenating the WHIN pre-training embedding with the representation obtained after processing the text information through BERT. All member text information is re-weighted through an attention mechanism based on the skills $s_i$ required by $j_0$. A neighbor sampler module, operating based on the similarity between different members and the required skills, can filter out professional connections that are not relevant to the job.
• Figure 4: Hyperparameter tuning experiments to investigate the specific effects of social relations and job-specific attention mechanism.
• Figure 5: A case where professional connections improve performance on Person-Job Fit. CSAGNN can improve the performance of Person Job Fit by filtering and aggregating information from professional networks. For privacy protection reasons, we have rewritten the statement in the example while ensuring that the semantic information remains unchanged.