Sage Journals: Discover world-class research

Abstract

Previous studies have suggested that emotions influence individuals’ passage of time judgments (PoTJs). However, it remains unclear how this relationship manifests in daily life. Weibo provides a wealth of daily verbal expressions from individuals, offering valuable data for analyzing how emotions and PoTJs vary. Utilizing this resource and choosing a specific period of time—COVID-19—the authors innovatively examined the relationships between emotional valence, discrete emotions, and PoTJs across different timescales: weekly, bi-weekly, monthly, and over the entire year. First, they constructed a PoTJ lexicon and template based on Weibo data. This was followed by an analysis of emotional expressions in PoTJ-related posts extracted by the lexicon and template using natural language processing techniques. The findings reveal a distinct pattern: most individuals perceived time as passing quickly during the COVID-19 pandemic, while fewer experienced slow PoTJ. Regarding the relationship between emotional valence and PoTJ, the daily average percentage of posts expressing positive and negative emotions was positively correlated with fast and slow PoTJs from a dynamic perspective, respectively, and a similar pattern was found from an overall perspective. More importantly, a unique relationship was identified between discrete emotions and PoTJs. Specifically, both the dynamic and overall analyses showed a robust finding: only joy and anxiety were positively associated with fast and slow PoTJs, respectively. In contrast, for other emotions (e.g., hope, sadness), the dynamic analyses yielded inconsistent results across the different time windows, diverging from the overall analysis. These findings highlight the importance of adopting a dynamic perspective and considering discrete emotions when understanding the relationship between emotions and PoTJs.

Keywords

Emotion passage of time judgment natural language processing Weibo dynamic perspective

Introduction

Passage of time judgment (PoTJ) refers to the perception of time passing, which plays a crucial role in daily life. Distortions in PoTJ are linked to various negative symptoms, including depression (Ogden, 2021), overeating (Isham et al., 2022), and drug use (Ogden & Faulkner, 2022). Given its significance, understanding the factors influencing PoTJ has attracted considerable attention from researchers (Brenlla et al., 2022; Droit-Volet & Dambrun, 2019; Droit-Volet & Wearden, 2016; Martinelli & Droit-Volet, 2022b; Ogden, 2020; Wearden, 2015), and emotion is a key factor (Droit-Volet et al., 2020; Liu et al., 2024; Martinelli et al., 2021). However, studies have yet to examine PoTJ and emotion from both an overall and a dynamic perspective using ecologically valid methods.

To investigate the relationship between emotion and PoTJ from both an overall and a dynamic perspective in daily life, this study focuses on the COVID-19 pandemic, a period during which emotional expression on social media underwent dramatic changes. During this time, negative emotion surged globally, followed by a gradual recovery (Wang et al., 2022). In China, negative emotion (e.g., fear) increased significantly in early 2020, followed by a rise in positive emotional expressions (i.e., praise words) later in the pandemic (Shi et al., 2022; Su et al., 2021). Changes in PoTJ were also reported, with some studies indicating that time passed faster, while others observed that it passed more slowly (Alatrany et al., 2022; Brenlla et al., 2022; Kosak et al., 2022; Ogden, 2020, 2021). Emotion is a key factor in PoTJ changes (Droit-Volet & Wearden, 2016; Martinelli & Droit-Volet, 2022a, 2022b), with negative emotions, such as fear and boredom, typically associated with slower PoTJ, while positive emotions like happiness and joy are linked to faster PoTJ (Martinelli et al., 2021). However, the relationship between emotion and PoTJ changes remains unclear, particularly regarding whether increases in negative emotion during the early stages of the pandemic were associated with a greater expression of slow PoTJ.

Although existing studies highlight a strong relationship between emotion and PoTJ (Droit-Volet et al., 2020; Liu et al., 2024; Ogden, 2020), many rely on survey data, which may introduce biases due to experimenters’ intentions or questionnaire content (Galesic & Bosnjak, 2009; Rockwood et al., 1997; Sjöström & Holst, 2002). In contrast, qualitative data from social media offers advantages in terms of ecological validity, sample size, and cost-efficiency (Andrews et al., 2015; Subramanian et al., 2023; Takayasu et al., 2015). Weibo, with over 224 million daily active users in 2020 (Sina Weibo Data Center, 2021), serves as a rich source of emotional expressions during the pandemic, revealing significant shifts in emotion (Han et al., 2022; Shi et al., 2022). Social media data surpasses traditional surveys in terms of volume, accessibility, and time span, making it an ideal tool for studying the dynamic relationship between emotion and PoTJ.

With advances in machine learning and natural language processing techniques, online texts from social platforms can be used to analyze individuals’ emotion or psychological characteristics (Nguyen & Van Nguyen, 2020; Zahoor & Rohilla, 2020). Recent research on sentiment analysis has found that machine learning algorithms trained with Weibo text data can accurately classify text into two or more emotion categories (Wu et al., 2018; Zhang et al., 2022). Word2vec, a statistical model, converts words into vectors and is used in psychology (Hakonen et al., 2022; Xu et al., 2014, 2021), allowing researchers to calculate the association between emotional words and posts, thus providing insights into emotional expressions on Weibo. Hence, the Weibo big-data-based approach offers a way to study the relationship between emotions and PoTJs with high ecological validity, and by analyzing the Weibo data set, we can explore the dynamic relationship between emotion and PoTJ.

How does the passage of time connect to emotion? And how does it vary with changes in emotions? To address these questions, this study employed the advanced pre-trained Bidirectional Encoder Representations from Transformers (BERT) model and Word2vec model to analyze the relationship between emotion and PoTJ in a large-scale Weibo data set. Initially, we developed a PoTJ lexicon and template to extract PoTJ-related Weibo posts, encompassing both fast and slow PoTJ posts. Subsequently, we constructed a sentiment classification model alongside a discrete emotion classifier to analyze emotional valence and the discrete emotion types present in these posts. Our overall analysis examined the relationship between emotion and PoTJ averaged over one year, while dynamic analyses investigated the correlation between monthly average emotion and PoTJ over the year. We further conducted similar correlation analyses using bi-weekly and weekly time windows to explore the stability of this relationship across different time frames. Through these analyses, we aim to enhance our understanding of the interplay between emotion and PoTJ.

Method

Data Acquisition

The PoTJ-Weibo data set was extracted from the Weibo-COV V2 data set (Hu et al., 2020), covering posts from January 20, 2020 to December 31, 2020, when COVID-19 was officially classified under the Law of the People's Republic of China on the Prevention and Control of Infectious Diseases. The data set ends on December 31, 2020, at 23:59 (Greenwich Mean Time + 8). The filtered Weibo data set contained 65,127,978 posts from 10,833,065 active users during the period (hereafter referred to as Weibo-Large). Preprocessing was performed using HarvestText to remove special characters, retaining only the user's reason for reposting. This study adhered to the Declaration of Helsinki and received approval from the Ethics Committee of the Department of Psychological and Cognitive Sciences at Tsinghua University.

PoTJ-Related Data-Set Extraction

Using the self-developed PoTJ lexicon and template (for details, see the Appendix), we extracted a PoTJ-related data set from the Weibo-Large corpus. Our goal was to identify posts that explicitly describe the subjective experience of time passing quickly (fast PoTJ) or slowly (slow PoTJ). Importantly, these labels were assigned based on explicit linguistic cues, not inferred sentiment, to ensure a transparent and replicable classification process. The extraction procedure consisted of three steps.

Lexicon-Based Extraction

Each post in the Weibo-Large corpus was segmented into words using the Jieba tool kit and matched against entries in the validated PoTJ lexicon, which was developed through participant-generated responses, semantic expansion, and expert selection (see the Appendix). Posts containing at least one matched term—for example, 时光飞逝 (“time flies”)—were retained as candidates.

Template-Based Extraction

In parallel, we applied a syntactic pattern-matching procedure based on the PoTJ template—for example, [时间]过[得]慢 (“[time] passed slowly”)—which was developed through a process similar to the lexicon. Each post in the Weibo-Large corpus was segmented into sentences based on punctuation and retained if at least one sentence matched a pattern of the template. To address the linguistic ambiguity of Chinese expressions (Huang, 2008), we excluded posts containing modal verbs (e.g., hope, wish) or words with lexically ambiguous morphemes (e.g., the character 快 “fast” in 快乐 “happiness”), based on a manually curated exclusion list.

Merging and Filtering

We merged the results from the lexicon- and template-based extractions and removed the duplicates. To exclude content such as news releases and advertisements that do not reflect personal time-passage experiences, we calculated pairwise similarity and removed posts with an over 80% word overlap. Additionally, posts containing both fast and slow PoTJ expressions were excluded to avoid interpretive ambiguity.

Through this process, we constructed the final PoTJ-Weibo data set, comprising 40,723 posts labeled as expressing fast PoTJ and 3,845 posts labeled as expressing slow PoTJ, drawn from the Weibo-COV V2 corpus containing 65,127,978 posts. These categorizations were based on the explicit semantics of the extracted expressions and reflect individuals’ spontaneous self-reports of subjective time experience in naturalistic online contexts.

Sentiment Analysis on PoTJ-Weibo Data Set

We conducted sentiment analysis to explore the relationship between emotion and PoTJ, focusing on both emotional valence and discrete emotions. Specifically, emotional valence classification was used to identify the overall affective polarity of each post (i.e., positive, negative, or neutral), while discrete emotion classification captured more specific emotional categories such as joy, sadness, or anger. These two types of classification rely on distinct emotional taxonomies and modeling techniques, as described below.

Emotional Valence Classification

The emotional valence classification of text is a classic task in sentiment analysis (Mohammad, 2016; Zhang et al., 2018). In this analysis, the valence of the emotion expressed in the post was classified into three types: positive, negative, and neutral. A deep neural network was utilized to conduct the emotional valence classification task based on large-scale data. We adopted a pretrained BERT model (Devlin et al., 2018) to encode each post into latent vectors, which were then passed through a trainable multi-layer perceptron, producing the emotional valence of the post. For the pretrained BERT model, a Chinese BERT wwm-ext model was utilized, taking a post as input and outputting a latent vector of 768 dimensions.¹

During the training stage, the parameters of the pretrained BERT model were fixed, and the multi-layer perceptron classifier was trained. The data set from The Evaluation of Weibo Emotion Classification Technology competition of The Ninth China National Conference on Social Media Processing, 2020 was adopted as the training and validation set, which includes more than 30,000 general posts and 13,000 COVID-19-related posts with sentiment annotations.² Its pandemic-related content closely aligns with the context of our study, making it a highly relevant foundation for training the emotional valence classification model. The data set labels six emotion categories, which we mapped into three emotional valence categories: happy and surprise were considered positive; anger, sadness, and fear were labeled negative; and neutral remained neutral. To classify the emotional valence of the COVID-19-related corpus, we used both all general posts and the training set of COVID-19-related posts as the final training set, and the test portion of COVID-19-related posts as the validation set. In total, the training and validation sets comprised 43,250 and 2,964 posts, respectively.

During the model training, cross-entropy loss was used as the loss function, and an early stopping strategy with 10 steps was adopted. Adam was used as the optimizer, and a two-layer perceptron with hidden dimensions of 1,024 and 512 was used as the predictor.

Discrete Emotion Classification

To capture more nuanced emotional expressions, we also conducted discrete emotion classification using Emo-Dict, a manually constructed lexicon of 3,156 Chinese emotion words categorized into eight types (Xu et al., 2021): love, joy, hope, surprise, disgust, sadness, anger, and anxiety. To determine the dominant discrete emotions in each post, we computed the similarity between post embeddings and Emo-Dict entries using word embedding vector similarity. This technique has been widely applied in natural language processing tasks such as query–answer matching (Shen et al., 2017) and sentence similarity assessment (Yao et al., 2018).

First, vector representations of all the words in Emo-Dict were obtained from the Word2Vec model trained in the section “Generation of PoTJ Lexicon”in Appendix where words not in Word2Vec were ignored, leaving vector representations of 2,902 emotional words corresponding to eight emotions in total. Here, we denote the j-th word for the k-th emotion in Emo-Dict as $d_{k, j}$ , and the vector of $d_{k, j}$ as $v_{d_{k, j}}$ .

Second, the sentiment scores for the eight emotions were estimated for each post in the PoTJ-Weibo data set. We split each post into words using Jieba, fetched vectors for all the words from Word2Vec, and calculated their similarity with each word in Emo-Dict:

s (ω_{i}, d_{k, j} = \frac{v_{ω_{i}} \cdot v_{d_{k, j}}}{‖ v_{ω_{i}} ‖ \cdot ‖ v_{d_{k, j}} ‖})

where

ω_{i}

indicates the i-th word in the post,

v_{ω_{i}}

indicates the vector representation of

ω_{i}

, and

‖ \cdot ‖

is the magnitude of the vector.

Third, as the semantic representation of multiple vocabularies in the same emotion may differ considerably and it is necessary to estimate the emotion most likely to be expressed by one word $ω_{i}$ , we took the maximal similarity of all the Emo-Dict words ${d_{k, 1}, \dots, d_{k, n_{k}}}$ for an emotion k to represent the emotion score of $ω_{i}$ on emotion k:

s (ω_{i}, k) = Max ({s (ω_{i}, d_{k, 1}), \dots, s (ω_{i}, d_{k, n_{k}})})

where

n_{k}

is the n-th word for emotion k in Emo-Dict.

Finally, we obtained the post emotion score $s_{k}$ by averaging the emotion scores $s (ω_{i}, k)$ of the words in the post:

s_{k} = \frac{1}{| E_{k} |} \sum_{k \in E_{k}} s (ω_{i}, k)

where

E_{k}

is the words in the post with a strong relation with emotion k—that is, words

ω_{i}

with

s (ω_{i}, k) > n t; θ

(

θ

is a predefined threshold). In the end, the emotion E of the post was represented with the emotion with the largest score,

E = \arg max_{k} s_{k} (k \in {1, 2, \dots, 8})

In practice, we utilized $θ = 0.8$ . For discrete emotion recognition, we omitted posts without words with strong emotions, leaving 38,869 posts (87.2%) in the PoTJ-Weibo data set for the analysis of discrete emotions.

Both the emotional valence classifier (BERT + multi-layer perceptron) and the discrete emotion classifier (Word2Vec + Emo-Dict) were chosen for their prior validation in large-scale Chinese social media research and their computational efficiency (Guo et al., 2021; Li et al., 2023; Xu et al., 2014, 2021). In addition, the emotional valence classifier was fine-tuned on our data set to better align with the current task. Nonetheless, we recognize that these models do not offer intrinsic interpretability or contextual adaptability, which may limit the transparency of the sentiment classification. We further elaborate on these limitations in the discussion below.

Data Analysis

We analyzed the variation in PoTJ and emotion throughout the year from both an overall and a dynamic perspective. Since the volume of the PoTJ-Weibo (fast) corpus is much larger than that of the PoTJ-Weibo (slow) corpus with over ten times the number of posts), we adopted a resampling-based balancing strategy inspired by the bootstrap method (Mooney et al., 1993) for a fair comparison: random sampling with replacement was performed on the PoTJ-Weibo (fast) corpus to select a subsample comparable in size to the PoTJ-Weibo (slow) corpus. Then, for both emotional valence and discrete emotions, using our classified labels, we calculated the ratio of the number of posts belonging to each emotion in the subsample of the PoTJ-Weibo (fast) and the entire PoTJ-Weibo (slow) corpus. If PoTJ is minimally influenced by emotional valence or specific discrete emotions, the ratio should approximate that observed under conditions of neutral emotion, as identified through the emotional valence classification task. In the analysis, we applied bootstrap sampling 1,000 times, generating a distribution of fast-to-slow ratios for each emotional valence and discrete emotion. Statistical tests were then performed on these 1,000 bootstrap-derived ratio estimates for each condition, treating each iteration as approximately independent for the purpose of parametric testing. Specifically, we conducted a one-way analysis of variance (for emotional valence) and independent-sample t-tests (for discrete emotions versus neutral) using JASP (19.1.0). This resampling-based approach allows for the use of parametric tests while preserving comparability between the unbalanced corpora.

We further explored the association between emotions and PoTJ over time from a dynamic perspective. To show the overall temporal trends in posts, the daily average percentage of posts (DAPoP) for each month in 2020 from the PoTJ-Weibo (fast and slow) data sets was calculated, and similar calculations were performed for posts with different emotions (i.e., emotional valence and discrete emotions). Subsequently, the DAPoP for slow/fast and different emotions for each month was calculated, and correlation analyses were applied between the DAPoP of the negative/positive valence and the fast/slow PoTJ. The correlations between the DAPoP of discrete emotions and fast/slow PoTJ were also analyzed. We used percentages instead of total numbers for the correlation analysis to minimize bias from fluctuations in the volume of posts. In order to understand the dynamic relationship between emotion and PoTJ, we analyzed the average daily post percentages over weekly and bi-weekly time windows. This analysis allowed us to explore how emotion and PoTJ are related across different time windows, offering a dynamic perspective. We chose Pearson correlations as an initial exploratory measure due to their interpretability and compatibility with our DAPoP-derived summary statistics. However, we acknowledge their limitations in capturing temporal dependencies. Percentage-based time series are inherently compositional, and adjacent time windows may exhibit autocorrelation. These statistical constraints are addressed in Subsection of Limitations and Future Directions.

Transparency and Openness

The data and code related to the data extraction and sentiment recognition models are available on an open-source platform.³ The statistical tests, plotting, and lexicon-related data can be accessed on another platform.⁴ This study's design and its analysis were not preregistered.

Results

Overall Perspective

Overview of PoTJ-Weibo Data Set

A total of 40,723 posts indicated time passing quickly, while 3,845 posts indicated time passing slowly, resulting in a fast-to-slow ratio of approximately 10:1. Although the lexicon-based fast-to-slow ratio was 22:4 and the template-based ratio was 166:548, the extracted ratio remained close to 10:1 (10:0.73 for lexicons, 10:1.25 for templates). This suggests that most people experienced time passing quickly during this period rather than slowly.

Emotional Valence and PoTJ

The sentiment analysis of the PoTJ-Weibo data set revealed a nearly equal distribution of positive, neutral, and negative emotions (Figure 1a). As shown in Figure 1b, negative emotions were linked to a higher proportion of slow PoTJ, while positive emotions were associated with faster PoTJ. The one-way analysis of variance based on bootstrap-derived data revealed a significant main effect of emotional valence, F(2,2997) = 315327.465, p < .001, η_p² = 0.995, 95% confidence interval (CI) for η_p² [0.995, 0.996], indicating substantial differences in PoTJ across positive, neutral, and negative emotional conditions. Post-hoc tests revealed significant differences in the ratio of fast-to-slow PoTJ posts across emotion categories. The ratio was highest for positive posts (1.587 ± 0.034), followed by neutral (0.896 ± 0.020), and lowest for negative posts (0.777 ± 0.017; positive > neutral > negative, ps < .001, Cohen's |ds|> 4.832).

Figure 1.

Results of Emotion and PoTJ from an Overall Perspective

Discrete Emotions and PoTJ

Discrete emotions were also analyzed, with joy, hope, and love comprising the majority (73.4%) of the data set (Figure 1a). Independent-sample t-tests were conducted on the bootstrap-derived ratio to compare each discrete emotion against the neutral baseline. As shown in Figure 1c and Table 1, posts expressing love, joy, hope, surprise, and anger were more likely to be associated with fast PoTJ, t(1998) values > 53.050, ps < .001, Cohen's ds > 2.372, while sadness and anxiety were linked to slow PoTJ, t(1998) values < −123.811, ps < .001, Cohen's ds < −5.537. No significant difference was found between disgust and neutral emotion in terms of fast or slow PoTJ, t(1998) = −1.016, p = .310, Cohen's d = −0.045, 95% CI for mean difference [−0.015, 0.005].

Table 1.

Emotion Distributions of All Posts in the PoTJ-Weibo Data Set.

Emotion type	Discrete emotion		Neutral		t(1998)	p	Cohen's d	95% CI for mean difference
Emotion type	M	SD	M	SD	t(1998)	p	Cohen's d	Lower	Upper
Love	1.364	0.048	0.896	0.020	287.038	< .001	12.837	0.465	0.471
Joy	1.423	0.040	0.896	0.020	374.745	< .001	16.759	0.524	0.530
Hope	1.243	0.029	0.896	0.020	310.920	< .001	13.905	0.345	0.350
Surprise	2.306	0.185	0.896	0.020	239.657	< .001	10.718	1.399	1.422
Disgust	0.891	0.156	0.896	0.020	−1.016	.310	−0.045	−0.015	0.005
Sadness	0.752	0.031	0.896	0.020	−123.811	< .001	−5.537	−0.146	−0.141
Anger	1.215	0.189	0.896	0.020	53.050	< .001	2.372	0.308	0.331
Anxiety	0.269	0.017	0.896	0.020	−760.815	< .001	−34.025	−0.629	−0.625

Dynamic Perspective

PoTJ-Weibo Data Set Statistics

Figure 2a shows the dynamic changes of DAPoP for fast PoTJ over time, with a notable surge in the first three months followed by a gradual increase. Conversely, the DAPoP for slow PoTJ dropped sharply in the first three months, then displayed a consistent decline throughout the year.

Figure 2.

Results of Emotion and PoTJ from a Dynamic Perspective.

Emotional Valence and PoTJ

Figure 2b shows the DAPoP for both negative and positive emotional valence over 12 months. Positive emotions sharply increased in February and remained stable until September, with a notable rise in the last two months of the year. In contrast, negative emotions declined in February and December, with slight fluctuations from February to November.

Pearson correlation analyses were conducted between the monthly mean DAPoP values of emotional valence (positive, negative) and PoTJ (fast, slow). As shown in Table 2, the DAPoP of positive emotions was positively correlated with fast PoTJ, r = .603, p = .038, 95% CI for r [.045, .874], and negatively with slow PoTJ, r = −.611, p = .035, 95% CI [−.877, −.056]. Conversely, the DAPoP of negative emotions was positively correlated with slow PoTJ, r = .678, p = .015, 95% CI [.170, .901], and negatively with fast PoTJ, r = −.646, p = .023, 95% CI [−.890, −.115]. These correlations were significant in the monthly window but not in the weekly or bi-weekly windows (ps > .05).

Table 2.

Dynamic Correlation Between the DAPoP of Fast and Slow PoTJ Posts and the DAPoP of Positive and Negative Posts for Each Time Window.

Time window	PoTJ	Positive	p	Negative	p
Week	Fast	.144	.317	−.058	.696
	Slow	−.145	.316	.058	.691
Bi-week	Fast	.222	.287	−.178	.394
	Slow	−.222	.286	.179	.392
Month	Fast	.603	.038*	−.646	.023*
	Slow	−.611	.035*	.678	.015*

*p < .05.

Discrete Emotions and PoTJ

Figure 2c illustrates the DAPoP for discrete emotions over 12 months, showing distinct patterns for each emotion. For instance, hope decreased sharply in the first four months, then recovered, while anxiety steadily declined in the first half of the year. Sadness increased in the first three months, followed by fluctuations.

We conducted Pearson correlation analyses between the mean DAPoP of each discrete emotion and the corresponding mean values of fast and slow PoTJ across three temporal resolutions (weekly, bi-weekly, monthly). As presented in Table 3, joy and anxiety exhibited strong correlations with PoTJ across all time windows. Joy was positively correlated with fast PoTJ and negatively with slow PoTJ (ps < .05), while anxiety showed the opposite pattern (ps < .001). Hope was negatively correlated with fast PoTJ and positively with slow PoTJ in the weekly and monthly windows (ps < .05), showing a similar trend in the bi-weekly window (p = .078). Anger showed similar patterns in the weekly and bi-weekly windows (ps < .05). Sadness was positively correlated with fast PoTJ, r = .749, p = .005, 95% CI [.307, .925], and negatively with slow PoTJ in the monthly window, r = −.751, p = .005, 95% CI [−.926, −.311]. Other correlations were not significant (ps > 0.05).

Table 3.

Dynamic Correlation Between the DAPoP of Fast and Slow PoTJ Posts and the DAPoP of Discrete Emotion Posts for Each Time Window.

Time window	PoTJ	Love	p	Joy	p	Hope	p	Surprise	p	Disgust	p	Sadness	p	Anger	p	Anxiety	p
Week	Fast	.077	.596	.355	.011*	−.333	.018*	.222	.122	−.176	.221	.080	.581	−.395	.005**	−.740	<.001***
Week	Slow	−.076	.602	−.356	.011*	.333	.018*	−.222	.122	.176	.221	−.080	.579	.395	.005**	.740	<.001***
Bi-week	Fast	−.355	.082	.523	.007**	−.359	.078	.273	.187	−.187	.371	.210	.313	−.417	.038*	−.794	<.001***
Bi-week	Slow	.356	.081	−.523	.007**	.359	.078	−.273	.187	.187	.371	−.211	.312	.416	.038*	.795	<.001***
Month	Fast	.451	.141	.930	<.001***	−.768	.004**	.321	.310	−.311	.324	.749	.005**	−.291	.359	−.953	<.001***
Month	Slow	−.437	.155	−.933	<.001***	.766	.004**	−.314	.320	.301	.342	−.751	.005**	.276	.386	.963	<.001***

***p < .001. **p < .01. *p < .05.

Discussion

This study examines the relationship between PoTJ and emotions during the COVID-19 pandemic using Weibo data from both an overall and a dynamic perspective. The PoTJ lexicon and template were developed to extract relevant posts, and sentiment analysis was conducted to identify emotions. The results showed that most posts indicated time passing quickly, with fewer expressing slow PoTJ. From both perspectives, negative emotions were linked to slow PoTJ, while positive emotions were associated with fast PoTJ. Specific emotions, such as joy and anxiety, were associated with fast and slow PoTJ, respectively. Other emotions showed inconsistent patterns in the dynamic analyses across different timescales compared to the overall analysis. These findings are discussed in detail below.

More Expressions of Fast PoTJ May Be Linked to Hope, Joy, and Love

Contrary to the initial hypothesis, it was found that, during this unique period, most individuals expressed that time was passing quickly. This unexpected result may be due to the fact that the majority of PoTJ-related expressions were associated with hope, joy, and love. This could also be explained by the cultural influence of Confucianism, Buddhism, and Taoism, which emphasize a positive outlook during times of hardship (Xie & Wong, 2021). Such a cultural background may have contributed to individuals expressing more positive emotions (i.e., praise words) during the pandemic (Su et al., 2021), offering a plausible explanation for the greater frequency of fast PoTJ expressions. Additionally, the results confirm the role of discrete emotions in shaping PoTJ, reinforcing previous findings that emotion significantly influences PoTJ (Droit-Volet et al., 2020; Liu et al., 2024). Positive emotions such as joy, love, hope, and surprise were found to be associated with faster PoTJ, while negative emotions such as sadness and anxiety were linked to slower PoTJ. These results suggest that PoTJ is closely tied to an individual's internal state (Droit-Volet, 2018; Droit-Volet & Dambrun, 2019). However, the influence of emotional valence on PoTJ is not always consistent. For instance, anger, a typically negative emotion, was associated with fast rather than slow PoTJ. This indicates that PoTJ cannot be fully explained by emotional valence alone, as specific emotions may correspond to different PoTJs.

Relationship Between Emotion and PoTJ from a Dynamic and an Overall Perspective

During this specific period, distinct dynamic patterns were observed in PoTJ: fast PoTJ increased steadily after an initial surge, while slow PoTJ declined and then stabilized. These changes were correlated with emotional valence and some discrete emotions. Two key findings emerged regarding discrete emotions: first, joy and anxiety showed strong correlations with PoTJ across the time windows from both a dynamic and an overall perspectives; second, hope, anger, and sadness exhibited discrepancies between these perspectives, highlighting the complexity of their relationship with PoTJ.

Relationship Between Emotional Valence and PoTJ Across Different Timescales

The association between emotional valence and PoTJ was significant from both the overall and dynamic perspectives. However, from the dynamic perspective, the association was only observed at the monthly level, suggesting that shorter time frames may not capture a stable relationship. Fast PoTJ changes were positively correlated with positive emotions and negatively correlated with negative emotions, while slow PoTJ showed the opposite pattern. These findings align with previous research (Droit-Volet & Wearden, 2016; Droit-Volet et al., 2020; Kosak et al., 2022; Liu et al., 2024; Martinelli & Droit-Volet, 2022a; Martinelli et al., 2021) and reinforce that PoTJ is not only a static, emotion-dependent phenomenon but also fluctuates in response to emotional states over time. Notably, no significant correlations were found in the weekly and bi-weekly time windows. This may be due to shorter time frames introducing noise and reducing the stability of the emotion–PoTJ relationship (Pearson, 1896). These findings suggest that the association may be inconsistent across timescales and subject to moderating factors such as time-window length. For individual-level data, previous studies have reported mixed results on the effect of emotion on PoTJ across timescales (Alatrany et al., 2022; Ogden, 2021). Emotional valence was found to predict PoTJ on a 30-minute scale but not a daily scale (Liu et al., 2024), suggesting that its influence is more prominent in shorter timescales, where emotions are more stable. However, as timescales increase, individual emotions may fluctuate (Stone et al., 2006), potentially canceling out their effects on PoTJ. In this study, “different time windows” refers to the data collection scale rather than the intrinsic timescale of PoTJ fluctuations. Even on short timescales, emotions on Weibo fluctuate dynamically (Yu et al., 2021), making it difficult to identify a dominant emotion state, which may weaken the emotion–PoTJ relationship. Therefore, we speculate that a longer timescale is necessary for capturing stable emotional influence on PoTJ when using collective data, such as from Weibo. These results indicate that PoTJ cannot be solely attributed to emotional valence. Other influencing factors must be considered in future research.

Relationship Between Discrete Emotions and PoTJ

This study also explored the relationship between specific discrete emotions and PoTJ. Notably, joy and anxiety exhibited a strong association with PoTJ from both perspectives. Joy was positively associated with fast PoTJ and negatively associated with slow PoTJ, whereas anxiety showed the opposite pattern. These findings align with previous studies (Droit-Volet et al., 2020; Liu et al., 2024; Martinelli et al., 2021; Ogden, 2020) and emphasize that PoTJ can be more accurately predicted when considering discrete emotions rather than relying solely on emotional valence. However, discrepancies between the overall and dynamic perspectives were observed for emotions such as hope, anger, and sadness. In the overall analysis, hope and anger were associated with fast PoTJ, while sadness was linked to slow PoTJ. In contrast, the dynamic analysis revealed that changes in hope and anger were correlated with changes in slow PoTJ, while sadness was linked to fast PoTJ. These inconsistencies underscore the complexity of the relationship between specific emotions and PoTJ, and the importance of considering the temporal dynamics of emotion and PoTJ. We speculate that these discrepancies may be due to various factors, including the use of different time windows in the dynamic analysis, which may introduce variability in the data and contribute to inconsistencies in the results. Furthermore, individual differences in time expectations may also play a role in modulating the relationship between emotions and PoTJ (Liu et al., 2024; Tanaka & Yotsumoto, 2017). Future research could explore the moderating effects of such expectations to further clarify the underlying mechanisms that may account for these inconsistencies.

Limitations and Future Directions

This study provides initial insights into the relationship between emotion and PoTJ using spontaneous self-reported data from social media. However, several limitations should be noted regarding the generalizability of the results. Most notably, the analysis was conducted on population-level Weibo data collected during the COVID-19 pandemic. As such, whether the observed emotion–PoTJ patterns generalize to more stable or routine contexts remains an open question. Future studies could apply the same methodology to other time periods to assess generalizability.

Second, the sentiment classification models (BERT + multi-layer perceptron for emotional valence and Word2Vec + Emo-Dict for discrete emotions) were selected for their computational efficiency and prior validation on Chinese Weibo data (Guo et al., 2021; Li et al., 2023; Xu et al., 2021; Xue et al., 2014). However, both approaches have limitations. The BERT model, while powerful, does not provide feature-level interpretability, making it difficult to trace how single tokens contribute to classification outcomes. The Word2Vec model relies on static word embeddings and thus cannot disambiguate context-dependent meanings or adapt to semantic shifts. To address these limitations, future work could explore natural language processing techniques with greater interpretability or contextual adaptability, such as SHapley Additive exPlanations or the Local Interpretable Model-Agnostic Explanations framework (Lundberg & Lee, 2017; Ribeiro et al., 2016), to enhance model transparency and reliability.

Finally, the use of proportion-based time series entails potential issues such as compositional constraints and temporal autocorrelation, which can inflate correlations or obscure true effects. Although we used Pearson correlations as an exploratory measure, future studies could adopt more advanced approaches, such as time-series regression models or compositional data methods (e.g., centered log-ratio transformation) to improve robustness.

Conclusion

This study developed a novel Chinese PoTJ lexicon and template using the Weibo database, providing valuable resources for future research. The findings revealed a distinct pattern: most individuals perceived time as passing quickly during the COVID-19 pandemic, while fewer experienced slow PoTJ. Both the overall and the dynamic analyses revealed that negative emotions were linked to slow PoTJ, while positive emotions were associated with fast PoTJ. Joy and anxiety were strongly correlated with PoTJ, with joy associated with fast PoTJ and anxiety with slow PoTJ across both perspectives. However, sadness, anger, and hope exhibited variability from the dynamic perspective compared to the overall perspective. These findings underscore the importance of considering discrete emotions and dynamic perspectives for a comprehensive understanding of the emotion–PoTJ relationship.

Supplemental Material

sj-docx-1-pac-10.1177_18344909251390027 - Supplemental material for The Covariation of Emotion and Passage of Time Judgments: Insights from Weibo

Supplemental material, sj-docx-1-pac-10.1177_18344909251390027 for The Covariation of Emotion and Passage of Time Judgments: Insights from Weibo by Yanci Liu, Jiayu Li, Peixuan Han, Feng Du, Siyu Ma, Min Zhang and Meihong Zheng in Journal of Pacific Rim Psychology

Footnotes

Acknowledgments

Yanci Liu and Jiayu Li contributed equally to the article. The authors would like to express their sincere gratitude to the participants and reviewers for their valuable contributions to this study.

ORCID iDs

Yanci Liu

Meihong Zheng

Ethical Approval and Informed Consent Statements

Approval was obtained from the Ethics Committee of the Department of Psychological and Cognitive Sciences at Tsinghua University. The procedures used in this study adhere to the tenets of the Declaration of Helsinki.

Funding

This work was supported by the Natural Science Foundation of China (Grant No. U21B2026) and Tsinghua University's “Future Social Designer” program. The funding organizations had no role in the development of the study design or the collection, analysis, and interpretation of the data.

Declaration of Conflicting Interests

The authors declared no potential conflicts of interest with respect to the research, authorship, and/or publication of this article.

Data Availability Statement

Supplemental Material

Supplemental material for this article is available online.

Notes

References

Alatrany

S. S. J.

Ogden

Falaiyah

A. M.

ALdrraji

A. S.

Alatrany

A. S. S.

(2022). The passage of time in Iraq during the COVID-19 pandemic. PLoS ONE, 17(4), Article e0266877. https://doi.org/10.1371/journal.pone.0266877

Andrews

Ellis

D. A.

Shaw

Piwek

(2015). Beyond self-report: Tools to compare estimated and real-world smartphone use. PLoS ONE, 10(10), Article e0139004. https://doi.org/10.1371/journal.pone.0139004

Brenlla

M. E.

Germano

Seivane

M. S.

da Lama

R. F.

Ogden

(2022). Experiences of distortions to the passage of time during the Argentinian COVID-19 pandemic. PLoS ONE, 17(3), Article e0266261. https://doi.org/10.1371/journal.pone.0266261

Devlin

Chang

M.-W.

Lee

Toutanova

(2018). BERT: Pre-training of deep bidirectional transformers for language understanding. arXiv. https://doi.org/10.48550/arXiv.1810.04805

Droit-Volet

(2018). Intertwined facets of subjective time. Current Directions in Psychological Science, 27(6), 422–428. https://doi.org/10.1177/0963721418779978

Droit-Volet

Dambrun

(2019). Awareness of the passage of time and self-consciousness: What do meditators report? PsyCh Journal, 8(1), 51–65. https://doi.org/10.1002/pchj.270

Droit-Volet

Gil

Martinelli

Andant

Clinchamps

Parreira

Rouffiac

Dambrun

Huguet

Dubuis

Pereira

Bouillon

J.-B.

Dutheil

(2020). Time and COVID-19 stress in the lockdown situation: Time free, “dying” of boredom and sadness. PLoS ONE, 15(8), Article e0236465. https://doi.org/10.1371/journal.pone.0236465

Droit-Volet

Wearden

(2016). Passage of time judgments are not duration judgments: Evidence from a study using experience sampling methodology. Frontiers in Psychology, 7(176). https://doi.org/10.3389/fpsyg.2016.00176

Galesic

Bosnjak

(2009). Effects of questionnaire length on participation and indicators of response quality in a web survey. Public Opinion Quarterly, 73(2), 349–360. https://doi.org/10.1093/poq/nfp031

10.

Guo

Lai

Xiang

Huang

(2021). Emotion classification of COVID-19 Chinese microblogs based on the emotion category description. In S. Li, M. Sun, Y. Liu, H. Wu, L. Kang, W. Che, S. He, & G Rao (Eds.), Chinese computational linguistics (CCL 2021) (pp. 61–76). Springer. https://doi.org/10.1007/978-3-030-84186-7_5

11.

Hakonen

Ikäheimonen

Hultèn

Kauttonen

Koskinen

Lin

F.-H.

Lowe

Sams

Jääskeläinen

I. P.

(2022). Processing of an audiobook in the human brain is shaped by cultural family background. Brain Sciences, 12(5), Article 649. https://doi.org/10.3390/brainsci12050649

12.

Han

Pan

Zhang

(2022). Developmental trend of subjective well-being of Weibo users during COVID-19: Online text analysis based on machine learning method. Frontiers in Psychology, 12, Article 779594. https://doi.org/10.3389/fpsyg.2021.779594

13.

Huang

Chen

Mao

X.-L.

(2020). Weibo-COV: A large-scale COVID-19 social media dataset from Weibo. In K. Verspoor, K. B. Cohen, M. Conway, B. de Bruijn, M. Dredze, R. Mihalcea, & B. Wallace (Eds.), Proceedings of the 1st Workshop on NLP for COVID-19 (Part 2) at EMNLP 2020. Association for Computational Linguistics. https://doi.org/10.18653/v1/2020.nlpcovid19-2.34

14.

Huang

F. L.

(2008). Disambiguating effectively Chinese polyphonic ambiguity based on unify approach. In 2008 International Conference on Machine Learning and Cybernetics (Vol. 6, pp. 3242–3246). https://doi.org/10.1109/ICMLC.2008.4620965

15.

Isham

E. A.

Lomayesva

Teng

(2022). Time estimation and passage of time judgment predict eating behaviors during COVID-19 lockdown. Frontiers in Psychology, 13, Article 961092. https://doi.org/10.3389/fpsyg.2022.961092

16.

Kosak

Schelhorn

Wittmann

(2022). The subjective experience of time during the pandemic in Germany: The big slowdown. PLoS ONE, 17(5), Article e0267709. https://doi.org/10.1371/journal.pone.0267709

17.

Huo

Wang

(2023). Research on Chinese emotion classification using BERT-RCNN-ATT. WSEAS Transactions on Communications, 22, 17–31. https://doi.org/10.37394/23204.2023.22.2

18.

Liu

Song

Zheng

(2024). Factors influencing passage of time judgment in individuals’ daily lives: Evidence from the experience sampling and diary methods. Psychological Research, 88(2), 466–475. https://doi.org/10.1007/s00426-023-01859-z

19.

Lundberg

S. M.

Lee

S.-I.

(2017). A unified approach to interpreting model predictions. In I. Guyon, U. V. Luxburg, S. Bengio, H. Wallach, R. Fergus, S. Vishwanathan, & R. Garnett (Eds.), Advances in neural information processing systems (Vol. 30). Curran Associates, Inc. https://proceedings.neurips.cc/paper/2017/file/8a20a8621978632d76c43dfd28b67767-Paper.pdf

20.

Martinelli

Droit-Volet

(2022a). Judgment of duration and passage of time in prospective and retrospective conditions and its predictors for short and long durations. Scientific Reports, 12(1), Article 22241. https://doi.org/10.1038/s41598-022-25913-9

21.

Martinelli

Droit-Volet

(2022b). What factors underlie our experience of the passage of time? Theoretical consequences. Psychological Research, 86(2), 522–530. https://doi.org/10.1007/s00426-021-01486-6

22.

Martinelli

Gil

Belletier

Chevalère

Dezecache

Huguet

Droit-Volet

(2021). Time and emotion during lockdown and the COVID-19 epidemic: Determinants of our experience of time? Frontiers in Psychology, 11, Article 616169. https://doi.org/10.3389/fpsyg.2020.616169

23.

Mohammad

S. M.

(2016). Sentiment analysis: Detecting valence, emotions, and other affectual states from text. In H. L. Meiselman (Ed.), Emotion measurement (pp. 201–237). Woodhead. https://doi.org/10.1016/B978-0-08-100508-8.00009-6

24.

Mooney

C. Z.

Mooney

C. F.

Mooney

C. L.

Duval

R. D.

Duvall

(1993). Bootstrapping: A nonparametric approach to statistical inference. Sage.

25.

Nguyen

K. P.-Q.

Van Nguyen

(2020, November). Exploiting Vietnamese social media characteristics for textual emotion recognition in Vietnamese. In 2020 International Conference on Asian Language Processing (IALP) (pp. 276–281). IEEE. https://doi.org/10.1109/IALP51396.2020.9310495

26.

Ogden

R. S.

(2020). The passage of time during the UK COVID-19 lockdown. PLoS ONE, 15(7), Article e0235871. https://doi.org/10.1371/journal.pone.0235871

27.

Ogden

R. S.

(2021). Distortions to the passage of time during England’s second national lockdown: A role for depression. PLoS ONE, 16(4), Article e0250412. https://doi.org/10.1371/journal.pone.0250412

28.

Ogden

R. S.

Faulkner

(2022). The influence of recreational drug use on experiences of the passage of time. Sucht, 68(2), 65–74. https://doi.org/10.1024/0939-5911/a000761

29.

Pearson

(1896). VII. Mathematical contributions to the theory of evolution.—III. Regression, heredity, and panmixia. Philosophical Transactions of the Royal Society A, 187, 253–318. https://doi.org/10.1098/rsta.1896.0007

30.

Ribeiro

M. T.

Singh

Guestrin

(2016). “Why should I trust you?”: Explaining the predictions of any classifier. In Proceedings of the 22nd ACM SIGKDD International Conference on Knowledge Discovery and Data Mining (pp. 1135–1144). https://doi.org/10.1145/2939672.2939778

31.

Rockwood

T. H.

Sangster

R. L.

Dillman

D. A.

(1997). The effect of response categories on questionnaire answers. Sociological Methods & Research, 26(1), 118–140. https://doi.org/10.1177/0049124197026001004

32.

Shen

Rong

Jiang

Peng

Tang

Xiong

(2017). Word embedding based correlation model for question/answer matching. Proceedings of the AAAI Conference on Artificial Intelligence, 31(1). https://doi.org/10.1609/aaai.v31i1.11002

33.

Shi

W. Z.

Zeng

Zhang

Tong

Shen

Liu

Shi

(2022). Online public opinion during the first epidemic wave of COVID-19 in China based on Weibo data. Humanities and Social Sciences Communications, 9, Article 159. https://doi.org/10.1057/s41599-022-01181-w

34.

Sina Weibo Data Center . (2021). Weibo 2020 user development report. https://Data.Weibo.Com/Report/ReportDetail?Id=456&display=0&retcode=6102

35.

Sjöström

Holst

(2002). Validity of a questionnaire survey: Response patterns in different subgroups and the effect of social desirability. Acta Odontologica Scandinavica, 60(3), 136–140. https://doi.org/10.1080/000163502753740133

36.

Stone

A. A.

Schwartz

J. E.

Schkade

Schwarz

Krueger

Kahneman

(2006). A population approach to the study of emotion: Diurnal rhythms of a working day examined with the day reconstruction method. Emotion, 6(1), 139–149. https://doi.org/10.1037/1528-3542.6.1.139

37.

Xue

Zhu

(2021). Public emotion responses during COVID-19 in China on social media: An observational study. Human Behavior and Emerging Technologies, 3(1), 127–136. https://doi.org/10.1002/hbe2.239

38.

Subramanian

, D e Moor

Fiedler

Koniuch

Janowski

(2023). Towards enhancing ecological validity in user studies: A systematic review of guidelines and implications for QoE research. Quality and User Experience, 8(1), Article 6. https://doi.org/10.1007/s41233-023-00059-2

39.

Takayasu

Sato

Sano

Yamada

Miura

Takayasu

(2015). Rumor diffusion and convergence during the 3.11 earthquake: A Twitter case study. PLoS ONE, 10(4), Article e0121443. https://doi.org/10.1371/journal.pone.0121443

40.

Tanaka

Yotsumoto

(2017). Passage of time judgments is relative to temporal expectation. Frontiers in Psychology, 8, Article 187. https://doi.org/10.3389/fpsyg.2017.00187

41.

Wang

Fan

Palacios

Chai

Guetta-Jeanrenaud

Obradovich

Zhou

Zheng

(2022). Global evidence of expressed sentiment alterations during the COVID-19 pandemic. Nature Human Behaviour, 6, 349–358. https://doi.org/10.1038/s41562-022-01312-y

42.

Wearden

J. H.

(2015). Passage of time judgements. Consciousness and Cognition, 38, 165–171. https://doi.org/10.1016/j.concog.2015.06.005

43.

Kang

Matsumoto

Yoshida

Kita

(2018). Emoticon-based emotion analysis for Weibo articles in sentence level. In M. Kaenampornpan, R. Malaka, D. Nguyen, & N. Schwind (Eds.), Multi-disciplinary trends in artificial intelligence (MIWAI 2018) (pp. 104–112). Springer. https://doi.org/10.1007/978-3-030-03014-8_9

44.

Xie

Wong

D. F. K.

(2021). Culturally sensitive conceptualization of resilience: A multidimensional model of Chinese resilience. Transcultural Psychiatry, 58(3), 323–334. https://doi.org/10.1177/1363461520951306

45.

Jiang

Sun

Wen

Shi

Sun

Qian

(2021). A novel emotion lexicon for Chinese emotional expression analysis on Weibo: Using grounded theory and semi-automatic methods. IEEE Access, 9, 92757–92768. https://doi.org/10.1109/ACCESS.2020.3009292

46.

Xue

Zhan

(2014, June). A study on sentiment computing and classification of Sina Weibo with word2vec. In 2014 IEEE International Congress on Big Data (pp. 358–363). IEEE. https://doi.org/10.1109/BigData.Congress.2014.59

47.

Yao

Liu

Zhang

(2018). A novel sentence similarity model with word embedding based on convolutional neural network. Concurrency and Computation, 30(23), Article e4415. https://doi.org/10.1002/cpe.4415

48.

Eisenman

Han

(2021). Temporal dynamics of public emotions during the COVID-19 pandemic at the epicenter of the outbreak: Sentiment analysis of Weibo posts from Wuhan. Journal of Medical Internet Research, 23(3), Article e27078. https://doi.org/10.2196/27078

49.

Zahoor

Rohilla

(2020, July). Twitter sentiment analysis using machine learning algorithms: A case study. In 2020 International Conference on Advances in Computing, Communication & Materials (ICACCM) (pp. 194–199). IEEE. https://doi.org/10.1109/ICACCM50413.2020.9213011

50.

Zhang

Wang

Liu

(2018). Deep learning for sentiment analysis: A survey. WIRES Data Mining and Knowledge Discovery, 8(4), Article e1253. https://doi.org/10.1002/widm.1253

51.

Zhang

Jiang

(2022, September). Weibo short-text sentiment classification algorithm on serial hybrid network. In 2022 7th International Conference on Intelligent Computing and Signal Processing (ICSP) (pp. 535–539). IEEE. https://doi.org/10.1109/ICSP54964.2022.9778753

Supplementary Material

Please find the following supplemental material available below.

For Open Access articles published under a Creative Commons License, all supplemental material carries the same license as the article it is associated with.

For non-Open Access articles published, all supplemental material carries a non-exclusive license, and permission requests for re-use of supplemental material or any part of supplemental material shall be sent directly to the copyright owner as specified in the copyright notice associated with the article.

0.00 MB

0.06 MB