荷兰 诺达思 面部表情分析系统 FaceReader

摘要 · 中文整理：儿童在成就情境中会体验到多种情绪，然而除焦虑之外的数学相关情绪研究仍显不足，对即将进入小学的幼儿尤其如此。本研究报告了 15 名幼儿园阶段儿童在为期 3 个月的教学实验中解答难度递增的算术故事题时，其面部所表现的六种基本离散成就情绪（喜悦/高兴、悲伤、惊讶、愤怒、恐惧和厌恶）的普遍性与强度，并考察这些外显情绪在教学实验的初期、中期和末期对教学环节结束时算术正确率的影响程度。通过运用 FaceReader9，教学环节开始阶段表达最为强烈的三种情绪为高兴/喜悦、悲伤和惊讶。采用函数型回归分析发现，这些表达出来的成就情绪能够预测教学环节结束时的算术正确率；然而，当把教学环节随时间推进的效应纳入模型后，高兴/喜悦与正确率、悲伤与正确率之间的关系不再显著，而惊讶与正确率之间的关系依然显著。本文讨论了这些显著与不显著模式的可能解释。本研究是厘清情绪如何作用于问题解决行为的关键第一步，有助于我们思考如何以既肯定儿童情绪体验、又能促进数学学习生成的方式，回应幼儿有时强烈但始终存在的情绪。

原文摘要：Children experience a variety of emotions in achievement settings. Yet, mathematics-related emotions other than anxiety are understudied, especially for young children entering primary school. The current study reports the prevalence and intensity of six basic, discrete achievement emotions (joy/happiness, sadness, surprise, anger, fear, and disgust) expressed on the faces of 15 kindergarten-aged children as they solved increasingly complex arithmetic story problems in a 3-month teaching experiment. We also examine how the extent to which the expressed emotions influenced arithmetic accuracy at the end of an instructional session at the beginning, middle, and end of the teaching experiment. Through the application of FaceReader9, the three most intensely expressed emotions at the launch of the instructional sessions were happiness/joy, sadness, and surprise. Using functional regressions, these expressed achievement emotions predicted arithmetic accuracy at the end of the instructional session. However, when the effect of session over time was added to the model, the relationship between happiness/joy and accuracy, as well as sadness and accuracy, became non-significant. In contrast, the relationship between surprise and accuracy remained significant. We discuss potential explanations for these patterns of significance and non-significance. This study serves as a critical first step in clarifying how emotions contribute to problem-solving behavior as we grapple with how to respond to the sometimes intense, but always present emotions of young learners in ways that are affirming, as well as mathematically productive and generative.

关键词：成就情绪、幼儿数学、算术、故事应用题、自动面部编码

研究目标：解决幼儿数学学习研究中除焦虑之外的成就情绪长期被忽视、且幼儿难以用语言自评情绪的问题，借助自动面部编码技术量化幼儿在算术故事题解题过程中的离散情绪强度，并检验这些情绪能否预测同一教学环节结束时的解题正确率。

主要结论：在幼儿出现非中性表情时，强度最高的三种成就情绪依次为高兴/喜悦、悲伤和惊讶。函数型回归显示三者均能预测教学环节结束时的算术正确率，且系数随教学时间进程发生符号翻转；但在纳入教学实验推进时间的效应后，高兴与悲伤的预测作用不再显著，唯有惊讶与正确率的关系保持显著。研究表明学习并不要求持续的喜悦或完全没有负性情绪，自动面部编码是考察幼儿成就情绪与短期学习成效关系的可行方法。

创新点：其一，把自动面部编码技术从实验室刺激范式推广到真实的一对一教学情境，解决了幼儿情绪自我报告不可靠的难题；其二，采用函数型回归而非单点均值回归，刻画了情绪效应随教学时间推进的连续动态变化及符号翻转现象；其三，通过纳入教学实验时间协变量，区分了真实情绪效应与随实验推进产生的熟悉度效应，识别出惊讶情绪的稳健预测作用。

后续建议：扩大样本量并纳入不同社会经济与文化背景的幼儿，提升函数型回归系数曲线估计的稳定性与结论的可推广性；补充教师行为编码与师幼互动质量指标，考察成就情绪的情境诱因与教师反应的调节作用；在数学之外扩展到阅读、科学等学科任务，检验成就情绪与学业表现关系的学科特异性；引入皮电、心率等生理指标与自动面部编码做多模态融合，弥补幼儿表情微弱或被遮挡时的数据缺失；延长追踪周期至幼小衔接阶段，考察早期成就情绪模式对后续数学成就与数学焦虑的纵向预测作用

APA 引文：Kutaka, T. S., Chernyavskiy, P., & Hofkens, T. (2025). Achievement emotions in kindergarten: The association of solution accuracy with discrete joy, sadness, and surprise. Frontiers in Psychology, 15, 1466345. https://doi.org/10.3389/fpsyg.2024.1466345

仪器引用：All instructional sessions were videotaped by a GoPro© camera (Grant # Blinded), coded for accuracy and problem-solving strategy sophistication (Kutaka et al., 2023; Grant # Blinded), and then analyzed in the current study by FaceReader9 software (Noldus FaceReader 9, 2022) to encode the intensity of discrete universal emotions exhibited by each child (Grant # Blinded).

摘要 · 中文整理：本研究旨在阐明面部表情对帕金森病早期识别与诊断的作用。18 名帕金森病患者与 18 名对照者完成了音节发音测试，其面部运动被记录并用 Noldus FaceReader 7.0 进行分析，该软件通过捕捉分布于眉毛、嘴唇与鼻部各关键点的运动来推断面部表情的变化。研究以不分音节、单音节、双音节与多音节数据并结合人口学变量训练树增强贝叶斯网络模型。不分音节、单音节、双音节与多音节训练模型的曲线下面积分别为 0.960、0.958 与 0.962。正性面部表情与帕金森病概率呈负相关，而悲伤、愤怒、恐惧、厌恶等负性表情呈正相关，双音节模型中存在例外。朗读单音节所产生的恐惧表情具有最强的预测效应，而朗读多音节所产生的厌恶表情预测效应最弱。因此，发音任务中获取的面部表情参数可支持帕金森病的无创筛查。

原文摘要：We aimed to clarify the influence of facial expressions on providing early recognition and diagnosis of Parkinson's disease (PD). Eighteen PD patients and eighteen controls performed syllabic pronunciation tests while their facial movements were recorded and analyzed with Noldus FaceReader 7.0, which captures the movement of points distributed on the eyebrows, lips and nose to infer changes of facial expression. Tree-augmented Bayesian network models were trained on unsegmented-syllabic, monosyllabic, dissyllabic and multisyllabic data together with demographic variables. The area under the curve of the unsegmented-syllabic, monosyllabic, dissyllabic, and multisyllabic training models was 0.960, 0.958, and 0.962, respectively. Positive facial expressions showed a negative correlation with the probability of PD, while negative expressions such as sadness, anger, fear and disgust showed a positive correlation, with exceptions in the disyllabic model. Fear expressions generated by reading monosyllables showed the strongest predictive effect, whereas disgust generated by reading multisyllables showed the weakest effect. Facial expression parameters acquired during pronunciation tasks can therefore support the non-invasive screening of Parkinson's disease.

关键词：帕金森病、面部表情、贝叶斯网络、树增强网络、预测模型、Noldus 面部表情分析系统

研究目标：解决帕金森病早期诊断依赖临床量表评估、主观性强且难以在基层大规模推广的问题，探索能否用发音任务中采集的面部表情参数结合贝叶斯推断，实现对帕金森病的无创、低成本辅助识别。

主要结论：用 Noldus FaceReader 7.0 从音节发音任务录像中提取的面部表情参数，结合人口学变量所构建的树增强贝叶斯网络模型对帕金森病具有很高的判别能力，各音节模型的曲线下面积均达到 0.958 至 0.962。正性表情与患病概率负相关，负性表情总体上与患病概率正相关（双音节条件下悲伤、恐惧、厌恶为例外）；其中朗读单音节诱发的恐惧表情预测效应最强，朗读多音节诱发的厌恶表情预测效应最弱。面部表情量化可作为帕金森病无创筛查的辅助手段。

创新点：其一，把面部表情量化与发音任务相结合，用可控的音节朗读主动诱发面部肌肉运动，使面具脸这一难以自发观察的体征得以标准化量化；其二，按单音节、双音节、多音节分层建模，揭示了不同音节任务在判别效能上的差异；其三，采用树增强贝叶斯网络而非黑箱分类器，既保持高判别性能又能输出可解释的条件概率关系，明确各类表情与患病概率的关联方向。

后续建议：扩大样本量并设置独立外部验证队列，当前每组 18 例的样本规模不足以支撑高曲线下面积结论的稳健性；纳入不同 Hoehn-Yahr 分期的患者，检验模型对早期未用药患者的识别能力及与病情严重度的剂量反应关系；与统一帕金森病评定量表等临床金标准评分做相关分析，明确面部表情参数的临床意义边界；补充语音声学特征、手部运动等多模态指标与面部表情融合，提升模型的判别与分期能力；开展基于普通手机摄像头的可行性验证，推动该方法向基层筛查与远程随访场景落地

APA 引文：Mouse, M., Gong, H., Liu, Y., Xu, F., Zou, X., Huang, M., & Yang, X. (2025). Changes in facial expressions can distinguish Parkinson's disease via Bayesian inference. Frontiers in Neurology, 16, 1533942. https://doi.org/10.3389/fneur.2025.1533942

仪器引用：Noldus FaceReader 7.0 software was used to identify the movement of relevant muscle groups during the phonation test. More specifically, it captures the movement of points distributed on the eyebrows, lips and nose, then to infer the changes of face expression.

摘要 · 中文整理：理解消费者对葡萄酒的情绪反应对于改进营销策略与产品开发至关重要，情绪在塑造消费者偏好方面起着关键作用。本研究使用 FaceReader 软件 9.0 版（荷兰瓦赫宁恩 Noldus Information Technology 公司）进行面部表情分析，考察不同类型杜罗河葡萄酒（白葡萄酒、红葡萄酒与波特酒）所诱发的情绪反应。共有 80 名被试品尝了六款酒样，其面部表情被记录并加以分析。FaceReader 对高兴、悲伤、愤怒、惊讶、恐惧和厌恶等情绪的强度进行了量化，同时评估了唤醒水平。研究通过主成分分析（PCA）识别基于情绪反应的模式与分组。结果表明，白葡萄酒因其酸度而引发较多悲伤情绪，红葡萄酒则与较低水平的悲伤以及更强的舒适感相关；波特酒诱发惊讶情绪，这可能源于其甜型与加强型的特性。此外，在所有酒类中女性被试的唤醒水平都持续高于男性被试。本研究揭示了各类葡萄酒各自不同的情绪画像，并提示性别等人口学因素会影响情绪反应。这些洞察可为精准营销提供依据，并通过使葡萄酒特性与情绪投入更好地匹配来提升消费者体验。

原文摘要：Understanding consumers' emotional responses to wine is essential for improving marketing strategies and product development. Emotions play a pivotal role in shaping consumer preferences. This study investigates the emotional reactions elicited by different types of Douro wines (white, red, and Port) through facial expression analysis using FaceReader software, version 9.0 (Noldus Information Technology, Wageningen, The Netherlands). A total of 80 participants tasted six wine samples, and their facial expressions were recorded and analyzed. FaceReader quantified the intensity of emotions such as happiness, sadness, anger, surprise, fear, and disgust. Arousal levels were also assessed. The results were analyzed through principal component analysis (PCA) to identify patterns and groupings based on emotional responses. White wines evoked more sadness due to their acidity, while red wines were associated with lower levels of sadness and greater comfort. Port wines elicited surprise, probably due to their sweet and fortified nature. Additionally, female participants showed consistently higher arousal levels than males across all wine types. The study highlights distinct emotional profiles for each type of wine and suggests that demographic factors, such as gender, influence emotional responses. These insights can inform targeted marketing and enhance the consumer experience through better alignment of wine characteristics with emotional engagement.

关键词：葡萄酒情绪、面部表情分析、FaceReader 面部表情分析系统、杜罗河葡萄酒、消费者行为、主成分分析

研究目标：解决葡萄酒感官评价长期以风味描述与喜好打分为主、缺乏对消费者内隐情绪反应客观刻画的问题，探究白葡萄酒、红葡萄酒与波特酒三类杜罗河产区葡萄酒各自诱发的情绪画像，以及性别等人口学因素对情绪反应的调节作用。

主要结论：三类杜罗河葡萄酒具有可区分的情绪画像：白葡萄酒因酸度较高而诱发更多悲伤情绪，红葡萄酒对应较低的悲伤水平与更强的舒适感，波特酒因甜型加强酒的特性诱发惊讶情绪；女性被试在所有酒类上的唤醒水平均持续高于男性。研究证明 FaceReader 可用于构建葡萄酒的情绪-感官画像，并为按人群细分的精准营销提供依据。

创新点：其一，首次将 FaceReader 9.0 系统性应用于杜罗河法定产区白葡萄酒、红葡萄酒与波特酒三类产品的横向情绪比较，建立了产区导向的情绪画像；其二，在情绪强度之外同时纳入唤醒度指标并揭示稳定的性别差异；其三，用主成分分析把多维情绪输出压缩为可视化的情绪空间，使不同酒种的情绪聚类关系一目了然。

后续建议：将面部表情数据与葡萄酒的理化指标（总酸、残糖、酒精度、多酚含量）建立定量关联，解释情绪差异的化学成因；补充自我报告情绪问卷与购买意愿测量，检验内隐面部情绪与外显偏好、实际购买行为的一致性；扩大被试的国别与饮酒经验跨度，检验情绪画像是否受文化背景与专业程度调节；引入眼动追踪与心率变异性等多模态指标，区分酒标视觉信息与真实口感对情绪的独立贡献；采用时间维度的动态情绪分析（如时间强度法）刻画从入口到余味阶段的情绪演变轨迹

APA 引文：Marques, C., & Vilela, A. (2024). FaceReader insights into the emotional response of Douro wines. Applied Sciences, 14(21), 10053. https://doi.org/10.3390/app142110053

仪器引用：This study investigates the emotional reactions elicited by different types of Douro wines (white, red, and Port) through facial expression analysis using FaceReader software, version 9.0 (Noldus Information Technology, Wageningen, The Netherlands). A total of 80 participants tasted six wine samples, and their facial expressions were recorded and analyzed. FaceReader quantified the intensity of emotions such as happiness, sadness, anger, surprise, fear, and disgust.

摘要 · 中文整理：既往研究已确立面部表情作为判定食品与饮料愉悦度（总体愉快程度）的客观评估手段的效用。本研究旨在验证先前研究的结论，说明由呈味物诱发的面部表情具备预测这些呈味物主观愉悦度评分的能力。研究用数码相机记录了 29 名 18 至 55 岁女性被试饮用代表五种基本味的不同浓度溶液时的面部表情。借助广泛使用的面部表情分析软件 FaceReader 对面部表情进行细致评估，识别出惊讶、高兴、害怕、中性、厌恶、悲伤与愤怒七种情绪，其分数介于 0 至 1 之间，是情绪强度的数值化体现。与此同时，被试用 −5（极不愉快）至 +5（极愉快）的量表对每种溶液的愉悦度打分。研究采用多元线性回归分析构建了主观愉悦度评分的预测模型，并用另外 20 名被试对 11 种额外呈味溶液的情绪分数检验模型效力。预测的愉悦度评分与实测评分高度吻合且一致性良好，说明即便更换软件、呈味刺激与被试群体，先前的研究结论依然成立。文中还讨论了本技术在利用面部表情预测食品饮料愉悦度方面的若干局限与实践意义。

原文摘要：Previous studies have established the utility of facial expressions as an objective assessment approach for determining the hedonics (overall pleasure) of food and beverages. This study endeavors to validate the conclusions drawn from preceding research, illustrating that facial expressions prompted by tastants possess the capacity to forecast the perceived hedonic ratings of these tastants. Facial expressions of 29 female participants, aged 18–55 years, were captured using a digital camera during their consumption of diverse concentrations of solutions representative of five basic tastes. Employing the widely employed facial expression analysis application FaceReader, the facial expressions were meticulously assessed, identifying seven emotions (surprise, happiness, scare, neutral, disgust, sadness, and anger) characterized by scores ranging from 0 to 1—a numerical manifestation of emotional intensity. Simultaneously, participants rated the hedonics of each solution, utilizing a scale spanning from −5 (extremely unpleasant) to +5 (extremely pleasant). Employing a multiple linear regression analysis, a predictive model for perceived hedonic ratings was devised. The model's efficacy was scrutinized by assessing emotion scores from 11 additional taste solutions, sampled from 20 other participants. The anticipated hedonic ratings demonstrated robust alignment and agreement with the observed ratings, underpinning the validity of earlier findings even when incorporating diverse software and taste stimuli across a varied participant base. We discuss some limitations and practical implications of our technique in predicting food and beverage hedonics using facial expressions.

关键词：面部表情、愉悦度评分、呈味物诱发的情绪、预测模型、客观评估

研究目标：解决食品与饮品愉悦度评价过度依赖主观自我报告、易受语言表达能力与社会期许偏差影响的问题，验证由呈味刺激诱发的面部表情能否作为客观指标定量预测感知愉悦度，并检验该预测关系在更换分析软件、刺激类型与被试群体后的稳健性。

主要结论：由 FaceReader 8.1 自动编码的七种情绪强度可通过多元线性回归有效预测被试对味觉溶液及饮品的感知愉悦度，其中高兴、中性、悲伤、惊讶为显著预测变量，训练模型调整后 R² 达 0.614；在 20 名独立被试与 11 种新刺激上的外部验证显示预测值与实测值相关系数高达 0.977，一致性良好。单帧图像与 2 秒时间窗的预测效果优于 4 秒和 6 秒窗，说明仅凭一张代表性面部图像即可较好推断愉悦度。

创新点：其一，用独立被试与全新刺激集对愉悦度预测方程进行了严格的外部验证，而非仅报告训练集拟合优度；其二，系统比较了单帧、2 秒、4 秒、6 秒四种分析时间窗对预测效力的影响，给出了最佳时间窗的实证依据；其三，提出对极端愉悦度值施加 1.6 倍调整系数以校正回归的向均值收缩效应，使预测直线斜率接近 1。

后续建议：扩大被试构成，纳入更多男性与不同年龄、不同文化背景的被试，检验预测方程的跨性别与跨文化普适性；增加真实复杂食品（而非单一味觉溶液）作为刺激，考察质构、香气、温度等多感官因素对面部情绪的贡献；将 FaceReader 输出的面部动作单元与效价、唤醒度纳入模型，考察比七类基本情绪更细粒度的预测变量；引入心率、皮电、眼动等生理指标做多模态融合，解决部分被试对高愉悦刺激表情反应微弱的问题；用非线性模型（如随机森林、支持向量回归）替代多元线性回归，检验能否进一步提升预测精度并缓解多重共线性

APA 引文：Matsufuji, Y., Ueji, K., & Yamamoto, T. (2023). Predicting perceived hedonic ratings through facial expressions of different drinks. Foods, 12(18), 3490. https://doi.org/10.3390/foods12183490

仪器引用：Facial expressions were analyzed by FaceReader (ver. 8.1; Noldus Information Technology, Wageningen, The Netherlands). FaceReader processes facial expressions frame-by-frame at 30 Hz and classifies them into seven emotions (neutral, happy, sad, angry, surprised, scared, and disgusted) with scores ranging from 0 (no visible emotion) to 1 (emotion fully present).