動態公司感知的多任務學習方法於人才流動預測

Abstract

人才流動預測對企業制定有效的人才管理策略至關重要。在高度流動的全球勞動力環境中，企業必須分析人才市場動態，並制定相應策略以維持競爭力。透過人才流動預測，組織可以制定資料驅動的人才管理策略。現有研究已採用深度學習方法和線上專業網絡資料來解決人才流動預測問題。然而，這些研究專注在單向人才流動，因此未能捕捉人才流動的相互依賴性。此外，現有研究針對公司概況(company profile)的設計方式很難揭示組織特徵與人才流動之間的關係。 本研究旨在利用LinkedIn資料，開發深度學習模型進行人才流動預測。我們設計了雙向人才流動預測模型，此模型基於多任務成對架構以學習雙向的人才流動模式，並同時預測多方向的人才流動。此外，我們從人才流動紀錄中擷取具有時間性的公司特徵。我們的模型在預測人才流動量和辨識主要競爭公司方面皆展現出更優異的表現。

Talent flow prediction is critical for organizations in formulating effective talent management strategies. In today’s highly mobile global workforce, companies must analyze talent market dynamics and develop appropriate strategies to maintain their competitiveness. By leveraging talent flow prediction, organizations can develop data-driven strategies for talent management. Existing studies have applied deep learning methods and OPN data to address the talent flow prediction problem. However, these studies focus on uni-directional talent flow, which fails to capture the interdependent nature of talent flows. Moreover, the design of the company profile makes it difficult to reveal the relationship between organizational characteristics and talent flow. This study aims to develop a deep learning model leveraging LinkedIn data for talent flow prediction. We design a bi-directional talent flow prediction model, based on multitask pair-wise architecture to learn the bi-directional flow patterns and predict talent flow of multiple directions simultaneously. In addition, we extract temporal company profile features from talent flow records. Our model achieves superior performance in predicting talent flow amount and identifying key competitor companies.