運用錄影面試動態表情結合深度學習預測臺灣國際產業移工之留任意願：以卷積神經網絡為工具

Abstract

臺灣於2018年3月進入高齡社會階段，65歲以上的老年人口超過全人口的14%，勞力短缺問題逐漸加劇，國際移工成為支撐臺灣勞動力的不可或缺的一環。然而，國際移工在抵達臺灣後常常面臨失聯或怠惰等問題，且這些問題隨著時間的推移變得更加嚴重。臺灣對於移工失聯的法律約束不夠完善且程序繁瑣。因此，人力顧問公司希望在面試階段能夠篩選出願意留任的國際移工，以確保雇主能夠維持穩定的留任率。在心理學領域，隨著電腦視覺（Computer Vision）與深度學習（Deep Learning）技術的成熟發展，科技與心理領域的跨學科研究越來越多。許多學者開始合作，利用視訊錄影影片辨識當事人的動態表情，進而推測其情緒甚至未來的行為。本研究深度學習技術，即卷積神經網絡（Convolutional Neural Network, CNN），進行實證研究。研究對象為81位個案派遣公司所派遣的菲律賓和越南國籍產業移工，透過電腦視覺技術收集國際產業移工在特定情境下回答問題時所展現的面部動態表情軌跡，並利用卷積神經網絡建立動態表情與留任意願之間的模型預測他們的留任意願，為臺灣的移工雇主和派遣公司提供了一個快速而具有預測力的決策輔助工具，幫助他們在招募和甄選過程中做出明智的選擇。Taiwan is facing a growing labor shortage and has increasingly turned to international migrant workers to address this issue. However, there have been challenges associated with these workers, including cases of disappearance and low motivation after their arrival. The complex legal constraints and procedures have hindered the effective resolution of these problems. Consequently, it has become crucial to identify migrant workers who have a strong intention to stay during the interview process.This research aims to tackle this issue through the application of artificial intelligence, video interviews, dynamic expressions, sentiment analysis, and intention-to-stay assessment. We employed Convolutional Neural Network (CNN) to analyze the dynamic facial expressions of 81 international migrant workers from various case companies, utilizing an asynchronous video interview system. By capturing these workers' dynamic facial expression trajectories as they answered questions in specific scenarios, we developed a CNN-based model to predict their willingness to stay. This model serves as a practical solution to Taiwan's labor shortage problem. The results indicate that computer vision technology can effectively predict the intention to stay of international migrant workers during the interview stage. This predictive decision support tool offers valuable assistance to employers and dispatch companies in Taiwan, enabling them to recruit and select workers more efficiently and effectively.In conclusion, this study highlights the potential of utilizing artificial intelligence and computer vision technology to address Taiwan's labor shortage problem by predicting the intention to stay of international migrant workers during the interview stage. This tool facilitates the efficient and effective recruitment and selection of international migrant workers in Taiwan, contributing to a more sustainable labor force.