Institute of Psychology, Chinese Academy of Sciences
Institutional Repository of Institute of Psychology, CASNot a member yet
28529 research outputs found
Sort by
Artificial Intelligence Is Stereotypically Linked More with Socially Dominant Groups in Natural Language
No full textDespite the increasingly important role that artificial intelligence (AI) plays in society, its social representation remains underexplored. This study reveals that AI is asymmetrically stereotyped, being more closely associated with socially dominant groups. This conclusion is based on these investigations of AI's associations along the warmth-competence dimensions of the stereotype content model, its connections with socially advantaged (e.g., men, young, rich, white) versus disadvantaged (e.g., women, old, poor, non-white) demographic groups, and its perceived impact on high- versus low-prestige occupations. Across four studies using language-based analyses with static word embeddings, BERT-based models, and GPT-4o, as well as human-participant experiment validation, it is found that i) AI is strongly associated with high competence but exhibits variability in the warmth dimension (Study 1); ii) AI is more closely linked to advantaged demographic groups (Study 2); iii) advantaged demographic groups are semantically closer to AI than their disadvantaged counterparts along the warmth-competence dimensions (Study 3); and iv) high-prestige occupations (rather than low-prestige ones) are associated more strongly with AI's benefits than with its threats. Together, these findings indicate that public perceptions of AI are systematically biased toward socially dominant groups, potentially reinforcing existing social inequalities and raising concerns about an emerging "AI divide.
Learning Evaluation Models From Large Language Models for Sequence Generation
No full textAutomatic evaluation of sequence generation, which has traditionally relied on metrics such as BLEU and ROUGE, often struggles to capture the semantic accuracy of generated text due to an overemphasis on n-gram overlap. A promising solution to this issue is the development of model-based metrics, such as BLEURT and COMET. However, these approaches are typically limited by the scarcity of labeled evaluation data, which is essential for training evaluation models. In this work, we address this challenge by proposing the Customized Sequence Evaluation Metric (CSEM), a three-stage model training method that leverages large language models to generate labeled data for metric development, eliminating the need for human-labeled data. Furthermore, we extend the capabilities of CSEM to support a range of evaluation types, including single-aspect, multi-aspect, reference-free, and reference-based evaluations. This flexibility allows for the customization of metrics to fit various real-world scenarios. Experimental results on the SummEval benchmark demonstrate that CSEM can effectively train an evaluation model without human-labeled data. Additional experiments in reinforcement learning and reranking show that metrics developed through CSEM outperform traditional evaluation metrics, leading to significant improvements in sequence quality, as assessed by both commonly used metrics and ChatGPT
Exploring the alterations in microstate dynamics during the migraine cycle and detecting pre-ictal phases
No full textBackgroundMicrostate analysis captures brief but critical fluctuations in brain activity, making it a powerful tool for exploring the cyclic nature of migraine. In this study, we aimed to investigate microstate features during different migraine phases and develop a classification model to identify the pre-ictal phase.MethodsFrom May 2023 to June 2024, we conducted a cross-sectional study with consecutive recruitment, collecting resting-state electroencephalography data from 174 individuals with migraine without aura and 50 healthy controls, followed by classification of migraine phases. Microstate features, Lempel-Ziv complexity, and sample entropy were compared across five groups. A model was developed to identify the pre-ictal phase and validated on a test set.ResultsMicrostate features, particularly for microstates A and B, exhibited dynamic changes across the migraine cycle. The duration of microstate A was significantly longer in the inter-ictal phase than in the pre-ictal phase, whereas microstate B showed prolonged duration in the pre-ictal phase compared to healthy controls and the post-ictal phase. Microstate A displayed reduced coverage in the pre-ictal phase, whereas microstate B had increased occurrence and coverage during the pre-ictal and ictal phases. Transition probabilities also varied significantly: the pre-ictal phase showed elevated transitions from microstates A, C, and D to B, and the post-ictal phase showed reduced transitions from C and D to A. A classification model based on these microstate features achieved an area under the receiver operating characteristic curve (AUROC) of 0.85 (0.73-0.95), an area under the precision-recall curve (AUPRC) of 0.83 (0.66-0.95), and an F1 score of 0.78 (0.62-0.90) in the training set; and an AUROC of 0.84 (0.69-0.97), an AUPRC of 0.86 (0.67-0.98), and an F1 score of 0.81 (0.65-0.93) in the test set, indicating robust performance in identifying the pre-ictal phase.ConclusionThrough the observation of cyclic alterations in the microstates of patients with migraine, we identified a reduction in microstate A and an enhancement in microstate B during the pre-ictal phase. These changes may indicate a heightened sensitivity to auditory stimuli and increased activity in the visual cortex, providing new insights into migraine pathophysiology. Our model effectively identified the pre-ictal phase, offering a promising approach for early intervention in migraine attacks.</p
Cooperation patterns in depression: Insights from a meta-analysis of responsive and expressive social interactive paradigms
No full textThe use of high-ecological validity social interactive paradigms to study social impairments in depression has burgeoned in recent 10 years. However, abnormal social behaviors in depression have exhibited substantial yet unexplained variations, ranging from reduced to increased cooperation. We proposed an attentional focus theory and introduced a novel classification of social interactive paradigms, distinguishing between responsive (attentional focus on others) and expressive (attentional focus on self) paradigms. This framework systematically elucidates the underlying reasons for reduced and increased cooperation in depression. Depression encompasses current depressive episodes, remitted depression, and subclinical depressive symptoms in undiagnosed individuals. We systematically searched Web of Science, PsycInfo, and PubMed and performed a three-level metaanalysis on 44 studies with 71 effect sizes. We found an overall significant negative association between depression and cooperation, and a significant moderation effect of the responsive versus expressive category. Depression was negatively associated with cooperation in responsive paradigms overall and in each individual responsive paradigm (Cyberball game, responder-role Ultimatum Game, Prisoner's Dilemma, and investor-role Trust Game). Depression was positively associated with cooperation in expressive paradigms overall and in the trustee-role Trust Game, and a trend of positive association in allocation tasks. In our multiple-moderator analysis, while controlling for other significant single moderators, the responsive versus expressive category remained significant, explaining 66.32 % of the total variations in effect sizes. Our study completed the halfmissing landscape of social dysfunctions in social interactive paradigms in depression, which may help practitioners develop different social rehabilitation strategies tailored to responsive and expressive social scenarios.</p
Symmetrical and asymmetrical distortions in time and numerosity perception induced by chunked stimuli
No full textThe (co)representation of time and numerosity has long been a topic of enduring interest. While a theory of magnitude (ATOM) posits that these dimensions are governed by a shared representational system, empirical findings offer both supporting and conflicting evidence. Previous challenging research has highlighted that time and numerosity perception can be distorted in opposite directions by explicitly introducing emotional or cognitive interference. However, it remains unclear whether time and numerosity can spontaneously dissociate during stimulus processing. To this end, we tested the time and numerosity distortions caused by different kinds of chunked stimuli, including collinearity, illusory contours (ICs), and biological motion (BM). The results showed that collinearity caused the same amount of overestimation for both time and numerosity, whereas ICs caused only numerosity underestimation and BM caused only time overestimation. Notably, no consistent correlations emerged between the magnitudes of temporal and numerical distortion across the three stimulus types. These findings suggest that time and numerosity perception can be symmetrically or asymmetrically distorted depending on the nature of chunked stimuli, providing converging evidence for partially dissociable representations of time and numerosity. The close relationship observed between these two dimensions may instead reflect shared constraints within a broader framework of information processing
Symmetrical and asymmetrical distortions in time and numerosity perception induced by chunked stimuli
No full textThe (co)representation of time and numerosity has long been a topic of enduring interest. While a theory of magnitude (ATOM) posits that these dimensions are governed by a shared representational system, empirical findings offer both supporting and conflicting evidence. Previous challenging research has highlighted that time and numerosity perception can be distorted in opposite directions by explicitly introducing emotional or cognitive interference. However, it remains unclear whether time and numerosity can spontaneously dissociate during stimulus processing. To this end, we tested the time and numerosity distortions caused by different kinds of chunked stimuli, including collinearity, illusory contours (ICs), and biological motion (BM). The results showed that collinearity caused the same amount of overestimation for both time and numerosity, whereas ICs caused only numerosity underestimation and BM caused only time overestimation. Notably, no consistent correlations emerged between the magnitudes of temporal and numerical distortion across the three stimulus types. These findings suggest that time and numerosity perception can be symmetrically or asymmetrically distorted depending on the nature of chunked stimuli, providing converging evidence for partially dissociable representations of time and numerosity. The close relationship observed between these two dimensions may instead reflect shared constraints within a broader framework of information processing.</p
A Study on the Behavioral, Feelings, and Cognitive Styles of Adolescents with Major Depressive Disorder in the Context of Social Exclusion
No full text社会排斥对青少年抑郁症患者社会功能损伤具有重要影响。本研究通过改编Cyberball 社会排斥范式，创新性的构建了一种可量化的互动指标，用以评估青少 年抑郁症患者在社交排斥中的客观行为倾向，并对青少年抑郁症患者在社会排斥 中行为倾向、主观感受和认知风格进行研究。
研究分为两部分：研究一招募了 48 名健康成人参与者，验证改编的 Cyberball 招手范式的可行性及行为指标的效度和重测信度；研究二纳入 96 名青少年抑郁 症患者和 56 名健康对照，比较两组在社会排斥情境中的主观感受（需求满足和 情绪体验）、行为倾向（互动频次）以及认知风格（归因风格）的差异，并探讨 上述指标之间的关系及其与抑郁症状严重程度的关联。研究一结果表明，在互动 排斥情景下，个体主观感受降低，且招手互动行为随试次显著下降，并在一个月 后的重测中表现出中等程度的稳定性，提示该任务范式能实现预期的研究目的， 并具有一定的稳定性。研究二发现相较于健康对照，抑郁症患者在不同排斥水平 下普遍报告了更强烈的负性情绪和需求受威胁感，在排斥情景中表现出更高的社 交回避行为倾向以及对被排斥的解释抱有更消极的认知风格特征。相关分析进一 步揭示了青少年抑郁症患者在排斥情境中的“负面感受—消极认知—回避行为” 恶性循环机制： 抑郁焦虑症状通过引发社交互动中的负面感受从而强化社交回 避行为，被动回避反应又反过来巩固其被排斥状态，加深其在社会互动中的主观 痛苦程度，消极认知则加剧了负面感受和回避行为之间的影响。抑郁症状的严重 程度与这一恶性循环的各环节均呈正相关。相较于健康对照，青少年抑郁症患者 在社会排斥呈现出“越难过越退缩，越退缩越难过”的特定模式。
上述发现表明，本研究所构建的社会排斥行为指标为量化个体社交行为提供 了新工具；研究也揭示了青少年抑郁症患者在社会互动中易陷入情绪与行为交互 恶化的机制，为理解青少年抑郁症的社会功能障碍提供了实证依据，并为针对该 人群的识别和干预提供了理论参考。</p
Research on the Emotional Mediation and Intervention in the Relationship between Neuroticism and Identity Fusion
No full text身份融合是个体与群体之间形成高度紧密认同的心理现象，个体在高度身份融合状态下，会将自身的自我认同与群体认同深度绑定，使群体目标与自身目标趋于一致。这一现象不仅影响个体的社会行为、决策模式，还可能在特定情境下强化群体凝聚力，增强群体成员间的相互支持，并在极端情况下促使个体为了群体利益做出高度投入甚至自我牺牲的行为。因此，对身份融合的深入研究不仅有助于理解个体与群体关系的动态变化，还对群体管理、社会干预、极端行为预防及心理健康干预等领域具有重要的现实意义。
尽管身份融合已被广泛研究，但目前在测量方法、心理机制及有效的干预策略方面仍存在诸多挑战。传统的身份融合测量方法主要依赖自我报告问卷，如身份融合度量量表，但这些方法较难在动态情境下精准捕捉身份融合的变化，且可能受到社会期望效应或回忆偏差的影响。此外，身份融合的形成及变化受到复杂的心理和情绪因素调节，个体特质（如神经质）、情绪状态（如正负性情绪）及社会环境均可能对身份融合产生深远影响。然而，目前关于个体情绪在身份融合过程中的作用机制尚未得到充分探讨，而有效的干预策略在身份融合调节中的作用也缺乏系统性的实证研究。
为弥补上述研究空白，本研究围绕身份融合的测量、心理机制及个性化干预，通过三项研究系统探讨了身份融合的预测、影响因素及干预策略，为身份融合的测量方法、理论建构及干预应用提供了新的实证支持。
研究一采用行为数据分析与机器学习建模，探索个体行为特征对身份融合的预测能力，作为传统自我报告法的补充。研究共招募 180 名参考者，在兵棋对抗 模拟任务中采集其行为数据，研究使用支持向量回归（SVR）、线性回归（LR）、 LGBM 回归、随机森林回归（RF）和 XGBoost 随机森林回归（XGBRF）等方法进行建模，结果表明线性回归（r = 0.61, R² = 0.23）具有最佳预测效果，验证了行为数据作为身份融合客观测量指标的有效性。此外，基于 SHAP 分析，发现战损相关特征（如“重型坦克_战损_平均伤害”）是预测身份融合的关键变量。研究结果拓展了身份融合的测量方法，并为军事训练、团队管理及社会行为分析提供了新的方向。
研究二采用中介效应分析，探讨了神经质对身份融合的影响机制，并验证了情绪在其中的中介作用。研究发现，神经质与身份融合显著负相关（r=-0.28, p<0.01），即高神经质个体的身份融合水平较低；神经质与正性情绪负相关（r=- 0.42, p<0.01），与负性情绪正相关（r=0.53, p<0.01），说明高神经质个体更容易体验负性情绪，而较少体验正性情绪；正性情绪在神经质与身份融合之间起部分中介作用（间接效应占 30.38%，p < 0.001），高神经质个体较少体验正性情绪，从而削弱对群体的归属感；负性情绪在神经质与身份融合之间起更显著的中介作用（间接效应占 48.11%，p < 0.001），即高神经质个体由于更高的负性情绪水平，进一步降低了身份融合。此外，研究探讨了负性情绪在极端群体身份融合中的作用，提出通过增加极端群体成员的负性情绪可以降低其对群体的认同感，从而减少极端行为的发生。这一发现为去极端化策略提供了新的理论依据。
研究三采用实验研究设计，通过两个子实验探讨情绪干预在身份融合调节中的作用，并评估个性化情绪图片的干预效果。实验一研究不同负性情绪强度（高负性、低负性、中性）对身份融合的影响，发现高负性情绪组的干预效果最显著 （t = 5.49, p < 0.001, Cohen’s d = 0.96）。实验二对比大语言模型生成的个性化情绪图片与传统情绪图片库的干预效果，发现个性化干预效果更强（F(1,73) = 7.78, p < 0.01）。进一步分析表明：大语言模型（Large Language Model，LLM）组的情绪状态和身份融合水平变化更显著（p < 0.01），而传统情绪图片库（Chinese Affective Picture System，CAPS）组的干预效果较差。方差分析结果表明，组别效应（F(1,73) = 7.78, p < 0.01）和时间效应（F(1,73) = 27.36, p < 0.001）均显著，表明个性化情绪干预比标准化情绪干预更稳定。
本研究通过行为数据分析、心理机制探讨及个性化情绪干预，系统研究了身份融合的预测、影响因素及干预策略，并为群体管理、社会心理干预及去极端化策略提供了新的实证支持。研究结果不仅拓展了身份融合的测量方法，还为人工智能的情绪干预提供了新思路。未来研究可进一步优化预测模型，提高干预手段的个性化适配性，并探索身份融合在不同文化背景下的动态变化，以提升其在现实社会中的应用价值。</p
An investigation on the synchronization-based analgesic effect of music and movements
No full text疼痛管理是神经科学和心理学领域的重要课题。疼痛不仅是一种生理现象，还涉及复杂的认知、情感及运动过程。现有研究表明，音乐和运动都能缓解疼痛，音乐能够通过转移注意力、调节情绪和激活奖赏系统来减轻疼痛，而运动则通过调节下行疼痛通路、免疫系统以及通过运动诱发的感觉输入来抑制疼痛传导。值得注意的是，人类天生具有跟随音乐节拍同步动作的倾向。在日常生活中，音乐与运动的联合十分常见，这种自然的行为模式可能蕴含独特的镇痛机制。然而，音乐与运动联合的镇痛效果及其作用机制尚未得到充分研究。本研究基于节律性感觉运动同步理论，探讨音乐-运动同步对疼痛调节的影响。
本研究通过四个实验系统考察了音乐与运动结合的镇痛作用及其神经机制。实验一共采集了 80 名被试的行为数据，验证了音乐-击鼓联合干预的镇痛效果，并与单独的音乐或运动干预进行了比较。结果显示，与单独的音乐或运动干预相比，音乐-击鼓联合干预显著降低了疼痛强度和不愉悦度评分，表明节律性运动和听觉刺激的结合能够增强镇痛效果。
实验二进一步探讨了不同的联合方式（同相同步、反相同步、异步）对镇痛效应的影响，每种条件纳入了 22 名被试，研究发现当听觉节奏与运动同步（即 同相同步）时，镇痛效果最为显著，而异步条件下的镇痛效果较弱。这表明同步性在疼痛调节中具有关键作用。
实验三收集了78名被试的脑电数据，利用激光诱发电位分析音乐-运动同步 对疼痛相关脑响应的影响。结果发现，同相同步条件下激光诱发的 N2 成分幅值 显著降低，且与疼痛强度评分呈显著的负相关，表明N2成分的幅值下降介导了同步和疼痛减轻之间的关系。这表明音乐-运动同步可能通过调节躯体感觉皮层和岛叶皮层的活动，从而减少疼痛感知。
实验四进一步通过自发脑电活动分析，探讨了音乐-运动同步对神经振荡的调节作用，采集了 20 名被试干预前后的静息态数据。结果显示，在音乐-运动同步后，顶枕区 γ 频段功率显著增加，α 频段功率维持的比较稳定。此外，非周期指数分析显示，顶枕区的皮层兴奋性显著增强，表明音乐-运动同步可能通过增强神经网络的连通性和兴奋性，促进疼痛缓解。
本研究表明，节律性音乐-运动同步能够显著增强镇痛效果，并揭示了其潜在的生理机制。这种节奏一致性能够创造一种“心流”状态，使个体沉浸其中，并体验到高度的专注和愉悦感。这种状态可能会放大音乐的镇痛效果，提高疼痛耐受性。并为基于节律同步的非侵入性疼痛管理策略提供了生理学证据。这些发现不仅拓展了对疼痛感知调节的理解，也为未来开发音乐与运动结合的镇痛干预提供了新的思路。</p
Effects of convolutional neural network visual experience on face processing
No full text人类的面孔加工机制长久以来受到研究者关注。一张面孔可以为人类提供包 含身份、情绪等相当丰富的社会信息，有助于人的生活。在漫长的进化过程中， 人类获取了对面孔不同于普通物体的加工能力。卷积神经网络（convolutional neural network, CNN）是一种由连续卷积层和大量非线性运算单元构成的模型， 经过近几年迅速发展，已经在物体识别、面孔识别领域达到了人类水平。由于其 来源于人类神经网络生物学特性的设计和自主的学习过程，研究者开始关注卷积 神经网络加工与人类加工的相似性，已有相关研究通过训练 CNN 对大量面孔数 据进行身份识别任务来模拟人类面孔加工。然而，借助 CNN 模型探究人类视觉 认知加工机制是一条漫长的道路，相对于专家式面孔识别经验，较少有人探讨非 标签式面孔经验对面孔加工的影响。
本文以三个研究回答上述问题。研究一关注面孔视觉经验和广泛视觉经验的 两个 CNN 在方向、情绪、身份分类等面孔加工任务的表现。结果表明，相较于 广泛经验 CNN，面孔经验 CNN 更不易觉察局部倒置线索。同时，面孔经验 CNN 在身份分类任务上的表现较差，在情绪分类任务中的表现与广泛经验 CNN 没有 显著差异。这个结果表明非专家式面孔经验相比客体经验有助于 CNN 面孔加工 中的情绪分类任务而非身份分类任务。研究二探究不同视觉经验 CNN 的高低空 间频率面孔加工。结果表明面孔经验能使得 CNN 在面孔加工中出现低频优势， 人类实验数据也出现低频优势，但模式不同。这个结果初步表明非专家式面孔经 验而非客体经验有助于 CNN 面孔的低频加工，所导致的低频偏好表现与人类存 在相似之处。研究三从距离表征的角度探究 CNN 面孔与类面孔的加工，结果发 现不同视觉经验 CNN 对面孔-非面孔存在不同表征空间，主要表现为类面孔与真 实面孔的表征距离不同。
以上结果表明 CNN 的加工会受到非标签式视觉经验的影响，为人类低频面 孔加工现象提供一个可能的进化角度的来源，即可能与面孔特异性的非专家式优 化目标有关。</p
Access Repository Dashboard
Do you manage Institutional Repository of Institute of Psychology, CAS? Access insider analytics, issue reports and manage access to outputs from your repository in the CORE Repository Dashboard!