人格特質和心理韌性可利用生理和言語訊號預測

Abstract

大五人格和心理韌性是個人特質中相當重要的面向，可以被廣泛運用於學術界和業界，例如人事選拔或臨床心理領域。至今為止，其測量方式多半仰賴自陳式量表，不僅耗時耗力，也容易受填答者的主觀意識而影響作答。因此，本研究旨在藉由人機互動中所記錄的言語訊號與心電圖的振幅強度（Electrocardiogram (ECG) amplitude）、心率變異分析（Heart Rate Variability, HRV）和膚電生理回饋（Galvanic Skin Response, GSR）等生理訊號來預測大五人格特質和心理韌性。我們搜集32位受試者所提取並陳述的六段記憶——涵蓋正負向和三個時期（國小、國高中、大學）——以探討正負向情境和時間的潛在干擾並同時紀錄其對應的生理與語音訊號。首先，我們進行各訊號和問卷中各向度的相關分析來萃取有用的特徵訊號，再利用機器學習技術於這些特徵訊號來建立能預測個人特徵高低的二元分類模型。結果顯示：聲音訊號和生理訊號都能夠有效地預測大五人格和心理韌性的程度；預測模型對於高分組與低分組的二元區分可達0.68~0.86的F1分數。此研究證明行為與生理資料能夠有效地預測性格與韌性等重要個人特質。

Big-five personality and resilience are important personality characteristics and targets of interest in academic and industrial domains, such as personnel selection and clinical psychology. So far, the measurements of these two personal characteristics mainly rely on self-reported questionnaires, which are tedious and prone to response biases. To address these issues, the present study explored the possibility of predicting one’s resilience and big-five personality traits using speech and physiological signals measured during human-robot interactions. The audio features were extracted using OpenSMILE; the word usage was categorized using Linguistic Inquiry and Word Count (LIWC). The physiological signals include Electrocardiogram (ECG) amplitude, Heart Rate Variability (HRV), and Galvanic Skin Response (GSR). We asked 32 participants to retrieve six memories—positive and negative memories across three different periods—to balance the influence of valence. In the meantime, we recorded their audio and physiological signals during memory interpretation. We first examine correlations between personality traits and behavioral data and then build binary prediction models for factors in the big-five model and resilience. Results suggest that the audio modality and the physiological signals can serve as effective methods for predicting personality and resilience. Best achieved F1-scores range from 0.68 to 0.86 depending on different traits. Our research confirmed that behavioral features could provide effective cues for recognizing personal traits. To our knowledge, this is the first research using both speech and physiological signals to predict resilience.