Watching 360 videos using Virtual Reality (VR) head-mounted displays (HMDs) provides interactive and immersive experiences, where videos can evoke different emotions. Existing emotion self-report techniques within VR however are either retrospective or interrupt the immersive experience. To address this, we introduce the Continuous Physiological and Behavioral Emotion Annotation Dataset for 360 Videos (CEAP-360VR). We conducted a controlled study (N=32) where participants used a Vive Pro Eye HMD to watch eight validated affective 360 video clips, and annotated their valence and arousal (V-A) continuously. We collected (a) behavioral (head and eye movements; pupillometry) signals (b) physiological (heart rate, skin temperature, electrodermal activity) responses (c) momentary emotion self-reports (d) within-VR discrete emotion ratings (e) motion sickness, presence, and workload. We show the consistency of continuous annotation trajectories and verify their mean V-A annotations. We find high consistency between viewed 360 video regions across subjects, with higher consistency for eye than head movements. We furthermore run baseline classification experiments, where Random Forest classifiers with 2s segments show good accuracies for subject-independent models: 66.80% (V) and 64.26% (A) for binary classification; 49.92% (V) and 52.20% (A) for 3-class classification. Our open dataset allows further experiments with continuous emotion self-reports collected in 360 VR environments, which can enable automatic assessment of immersive Quality of Experience (QoE) and momentary affective states.

Recognizing user emotions while they watch short-form videos anytime and anywhere is essential for facilitating video content customization and personalization. However, most works either classify a single emotion per video stimuli, or are restricted to static, desktop environments. To address this, we propose a correlation-based emotion recognition algorithm (CorrNet) to recognize the valence and arousal (V-A) of each instance (fine-grained segment of signals) using only wearable, physiological signals (e.g., electrodermal activity, heart rate). CorrNet takes advantage of features both inside each instance (intra-modality features) and between different instances for the same video stimuli (correlation-based features). We first test our approach on an indoor-desktop affect dataset (CASE), and thereafter on an outdoor-mobile affect dataset (MERCA) which we collected using a smart wristband and wearable eyetracker. Results show that for subject-independent binary classification (high-low), CorrNet yields promising recognition accuracies: $76.37\%$ and $74.03\%$ for V-A on CASE, and $70.29\%$ and $68.15\%$ for V-A on MERCA. Our findings show: (1) instance segment lengths between 1–4 s result in highest recognition accuracies (2) accuracies between laboratory-grade and wearable sensors are comparable, even under low sampling rates (≤64 Hz) (3) large amounts of neutral V-A labels, an artifact of continuous affect annotation, result in varied recognition performance.

The advent of Artificial Intelligence (AI) technologies has precipitated the rise of asynchronous video interviews (AVIs) as an alternative to conventional job interviews. These one-way video interviews are conducted online and can be analyzed using AI algorithms to automate and speed up the selection procedure. In particular, the swift advancement of Large Language Models (LLMs) has significantly decreased the cost and technical barrier to developing AI systems for automatic personality and interview performance evaluation. However, the generative and task-unspecific nature of LLMs might pose potential risks and biases when evaluating humans based on their AVI responses. In this study, we conducted a comprehensive evaluation of the validity, reliability, fairness, and rating patterns of two widely-used LLMs, GPT-3.5 and GPT-4, in assessing personality and interview performance from an AVI. We compared the personality and interview performance ratings of the LLMs with the ratings from a task-specific AI model and human annotators using simulated AVI responses of 685 participants. The results show that LLMs can achieve similar or even better zero-shot validity compared with the task-specific AI model when predicting personality traits. The verbal explanations for predicting personality traits generated by LLMs are interpretable by the personality items that are designed according to psychological theories. However, LLMs also suffered from uneven performance across different traits, insufficient test-retest reliability, and the emergence of certain biases. Thus, it is necessary to exercise caution when applying LLMs for human-related application scenarios, especially for significant decisions such as employment.

During group interactions, we react and modulate our emotions and behaviour to the group through phenomena including emotion contagion and physiological synchrony. Previous work on emotion recognition through video/image has shown that group context information improves the classification performance. However, when using physiological data, literature mostly focuses on intrapersonal models that leave-out group information, while interpersonal models are unexplored. This paper introduces a new interpersonal Weighted Group Synchrony approach, which relies on Electrodermal Activity (EDA) and Heart-Rate Variability (HRV). We perform an analysis of synchrony metrics applied across diverse data representations (EDA and HRV morphology and features, recurrence plot, spectrogram), to identify which metrics and modalities better characterise physiological synchrony for emotion recognition. We explored two datasets (AMIGOS and K-EmoCon), covering different group sizes (4 vs dyad) and group-based activities (video-watching vs conversation). The experimental results show that integrating group information improves arousal and valence classification, across all datasets, with the exception of K-EmoCon on valence. The proposed method was able to attain mean M-F1 of $\approx$ 72.15% arousal and 81.16% valence for AMIGOS, and M-F1 of $\approx$ 52.63% arousal, 65.09% valence for K-EmoCon, surpassing previous work results for K-EmoCon on arousal, and providing a new baseline on AMIGOS for long-videos.

Fine-grained emotion recognition can model the temporal dynamics of emotions. It is temporally more precise when compared to predicting one emotion for activities (e.g., video clip watching). Previous works require large amounts of continuously annotated data to train an accurate recognition model. However, the experiments to collect large amounts of continuously annotated physiological signals are costly and time-consuming. To overcome this challenge, we propose a few-shot learning algorithm EmoDSN which can rapidly converge on a small amount of training data (typically < 10 samples per class (i.e., < 10 shot)) for fine-grained emotion recognition. EmoDSN recognizes fine-grained valence and arousal (V-A) labels by maximizing the distance metric between signal segments with different V-A labels. We tested EmoDSN on three different datasets, CASE, MERCA and CEAP-360VR, collected in three different environments: desktop, mobile and HMD-based virtual reality, respectively. The results from our experiments show that EmoDSN achieves promising results for both one-dimension binary (high/low V-A, 1D-2C) and two-dimensional 5-class (four quadrants of V-A space + neutral, 2D-5C) classification. We get an averaged accuracy of 76.04%, 76.62% and 57.62% for 1D-2C valence, 1D-2C arousal and 2D-5C respectively by using only 5 shot of training data. We also find that EmoDSN can achieve better recognition results trained with fewer annotated samples if we select training samples from the changing points of emotion and the ending moments of video watching.