Undergraduate students are in the critical period of developing self-identity; they face many challenges in academics, socializing, career planning, and intimate relationships, and their emotions are changeable and unstable. Recent research has indicated that emotional stability has significant predictive power over mental health. Emotional instability occurs in a wide range of mental disorders (Patel et al., 2015). In contrast, individuals with stable emotions reported a high degree of experienced happiness and life satisfaction (Hills and Argyle, 2001; Bajaj et al., 2019). Therefore, emotional stability is of great importance when considering improving mental health.

Individuals’ emotions change with their daily lives and interactions, and this has a biological basis. The suprachiasmatic nucleus regulates the circadian rhythm. The suprachiasmatic nucleus maintains the internal rhythm and keeps the same pace as the outer environment; this is done by receiving photic and non-photic cues and integrating them with the internal status. When suprachiasmatic nucleus rhythms are perturbed by genetic or environmental disturbances, brain regions related to mood regulation may be disrupted, leading to emotional disorders (Vadnie and McClung, 2017). In addition to the suprachiasmatic nucleus, individuals’ emotions may be disturbed by the circadian rhythm through the immune system, monoamine transportation, metabolic meditation, neuron regeneration, and other systems (McClung, 2013).

The daily rhythms of emotion are also associated with individuals’ daily routines and habits. Golder and Macy (2011) found that positive emotions rose in the mornings and decreased after midnight, while negative emotions remained high throughout the night. Moreover, for “night owls,” who slept during the day and were awake at night, their daily rhythms of emotions were different from those people with ordinary routines. Wang et al.’s (2016) study showed similar results. Bellivier et al. (2015) also found that almost all individuals with emotional disorders have an obvious daily rhythmic disturbance. Meanwhile, they found that disturbed circadian rhythms are closely associated with emotional fluctuations and disorders. Emotional fluctuations and disorders are more likely to emerge or further deepen in those who undertake shift work, regularly travel across time zones and have irregular habits. It is reasonable to infer that text materials reflecting individuals’ daily routines and life states are appropriate resources for understanding and analyzing individuals’ daily rhythms of emotions.

The dynamic assessment method is most commonly used to assess the daily rhythms of emotion; ecological momentary assessment, experience sampling method, diary methods (Bolger et al., 2003), and day reconstruction method are some examples of dynamic assessment methods. In a previous study, using these methods, participants were asked to assess their emotional states at different time points; thus, the trajectory of an individual’s emotion with changes in time and situation was drawn (Zhang et al., 2016). Specifically, using an ecological momentary assessment, participants repeatedly recorded their mental state and behavior in a natural environment with portable devices; thus, the researchers obtained data about individuals’ states in real-time (Shiffman et al., 2008). As there is no recall bias in the ecological momentary assessment, it has high ecological validity. In a study using the day reconstruction method, participants were asked to recall the previous day’s activities through questionnaires and assess the emotional states and experiences associated with the activities. As the activities had a timestamp, individuals’ daily rhythms of emotion, state, and other aspects were obtained (Kahneman et al., 2004).

However, in the methods mentioned above, individuals reacted to the assessment itself, which may lead to reactive bias. With the development of big data technology and social networks, researchers have begun to study individuals’ rhythms of emotion by analyzing big data from social networks (Wang et al., 2016; Eichstaedt and Weidman, 2020).

The data formed on social networking sites are naturally occurring and abundant. They provide a new, non-intrusive method to understand users’ mental states (Birnbaum, 2004) in which defensive thinking, social desirability effects, and memory bias can be avoided. In a study that included data from undergraduate students in the United States and Germany, Back et al. (2010b) found that data from social networks reflected participants’ mental states and personality traits. Further, data from social networks are tagged with time-stamps, making it possible to monitor and retrospectively examine the progress of mental state over time. Back et al. (2010a) explored the influence of terrorist attacks on people’s emotions by analyzing text content in their mobile phone messages before and after 9/11. Since text messages are time-stamped, emotional changes over time can be analyzed retrospectively.

A common way of excavating and analyzing data from social networks is to extract the features of text based on word frequency statistics and emotional tendency analysis (Cao et al., 2018) which were widely used in psychological studies (Liu et al., 2018; Sterling et al., 2020). This method is based on mental dictionaries. The most widely used mental dictionary is the Linguistic Inquiry and Word Count dictionary, which is appropriate for English text (Tausczik and Pennebaker, 2010). The dictionaries appropriate for analyzing simplified Chinese text are those in the Chinese mental analysis system (hereinafter called the “Wen Xin”) (Zhu, 2016) and HowNet (You et al., 2013). HowNet includes only positive and negative emotions. Alternatively, the dictionary in the “Wen Xin” system includes a fairly large number of categories of emotions; however, it was constructed based on the data from micro-blog specifically, and therefore, may not be appropriate for data analysis for other social networks.

Some researchers doubt the accuracy of text analysis based on word frequency and emotional tendency analysis, as text analysis does not consider irony, sarcasm, or unusual sentence structures (Pennebaker et al., 2003). However, Kahn et al. (2007) insisted on the validity of text analysis. According to their experiments, there was a significant correlation between the number of emotion words in a text based on the Linguistic Inquiry and Word Count dictionary and ratings for emotion by judges who analyzed the same text. This result indicates that text analysis, based on word frequency, is a valid reflection of an individual’s emotional state. Further, although irony and sarcasm may impact a specific sentence, they may offset each other when analyzing massive texts. Another study in China found that the frequency of a specific emotion word in a text was highly correlated with the human evaluation of emotion expression in that text (Zhao et al., 2016). The researchers also found that the efficiency of the Simple Chinese Linguistic Inquiry and Word Count dictionary in classifying emotions differed based on the type of text, with higher efficiency in news comments and RenRen Blog text and lower efficiency in micro-blog text. In general, text analysis based on the frequency of emotion words is valid for studying individuals’ emotions. It is worth noting that emotion dictionaries should be appropriated for different types of text.

For undergraduates in China, the most frequently used social networking site is Tencent QQ, which is similar to Facebook. Due to its multifunctionality, diversity, and personalized customization, Tencent QQ has become one of the most popular social networks among undergraduates (Zhang and Li, 2009). Tencent tweets (TTs), which are similar to the Facebook timeline, are the main way Chinese undergraduates post about their life situations and express their emotions and feelings. Therefore, the present study used the text analysis method to analyze text from TTs and explore the daily rhythms of undergraduate students’ emotions.

Undergraduate students’ emotions are influenced by demographic characteristics. Those who were once left behind in childhood had a higher rate of being diagnosed with depression (Han et al., 2017). Moreover, students from divorced families showed significantly higher levels of depression and anxiety (Huang and Li, 2011). Furthermore, female undergraduates showed more anxiety than male undergraduates, and those who reported high economic pressure had more emotional disorders (Stallman, 2010). Therefore, the other purpose of the present study is to examine whether there are differences in the daily rhythms of emotion among different demographic groups.

The present study explores the daily rhythm of undergraduate students’ emotions by analyzing TTs. This not only enriches the emotional analysis research of social media texts other than Twitter, Facebook, or micro-blog, but also provides new evidence for undergraduate students’ emotion expression on social networks within Chinese collectivist culture. This study has the potential to help students improve their emotional regulation abilities.

Many studies have been conducted on undergraduate students’ emotions; however, most investigated the states of their emotions rather than emotion fluctuation (Chen et al., 2016; Chester et al., 2016; Gloria and Steinhardt, 2016). In fact, emotional stability is important for individuals’ mental health. Researchers have examined dynamic changes in emotion based on social networks, using big data technology. Golder and Macy (2011) analyzed people’s diurnal and seasonal emotion rhythms using data from Twitter messages. Moreover, Wang et al. (2016) investigated emotion changes through data from Sina micro-blog, a popular social network in China. Instead of Twitter or micro-blog, Tencent QQ is the most widely used social network among Chinese undergraduates. They share their emotions and mental states with friends through TTs in Tencent QQ. Therefore, the present study analyzed TTs using big data technology to achieve three aims: (1) to determine whether TTs are suitable for the analysis of undergraduate students’ emotions; (2) to provide advice about emotion regulation strategies to undergraduates by exploring the status and rhythm of their emotions; and (3) to study intragroup differences in emotion status and rhythms.

Our results exhibited that the frequency of positive emotion words in TTs was much higher compared to negative emotion words. This finding is consistent with a general positive bias found in language; that is, the frequency of positive words is higher than that of negative words in both oral and written language (Dodds et al., 2015). However, there are more words to express negative emotions, both in existing dictionaries and the emotion dictionary constructed in this study. The frequency of posts about negative events in life was much lower than that of positive events, and thus the frequency of negative emotion words in the TTs was lower. Another possible reason for the lower frequency of negative emotion words is impression management; that is, undergraduates try to make a good impression on others by publishing more positive information on social networks (Xin et al., 2016).

In the positive emotion subcategories, the word frequency of love was much higher than that of other positive emotions, which is contrary to certain aspects of Chinese culture. Unlike the individualistic culture that encourages expressing feelings, the collectivist culture in China usually suppresses the free expression of emotions. That is to say, the Chinese are reluctant to express their negative feelings as well as one particular positive emotion—love. Previous studies have confirmed these findings. For example, compared to people from other cultures, Chinese people have been found to use emotion inhibition strategies more frequently and express fewer positive and negative emotions in emotional heuristic tasks (Soto et al., 2005; Wei et al., 2013). Moreover, they have been found to avoid expressing love (Caldwell-Harris et al., 2013). As far as we know, there are two explanations for the high word frequency of love in the TTs: the idealized virtual-identity hypothesis and the extended real-life hypothesis (Yao et al., 2014). The former holds that people tend to present an ideal self on social media; therefore, individuals are more likely to express positive emotions (Qiu et al., 2012). To create a good self-image, individuals are likely to express more love, which is a typical positive emotion, in TTs. The latter considers that social networks are an extension of real-life situations, and individuals will express their real selves on social media. Although Chinese individuals tend to be poor at expressing love in real situations, they do feel love and are more likely to express it on social media.

This study also found that undergraduates express depression, anxiety, and anger more than fear, which corresponds with their physical and mental characteristics and was consistent with the results of previous researches (Song and Fu, 2014; Cheng and Jia, 2019). Undergraduates are in the transition stage from adolescence to adulthood. In this important stage, they need to form a self-identity and face academic, interpersonal, emotional, economic, and other pressures. When individuals are unable to deal with these pressures, they are likely to experience depression and anxiety. The high word frequency of anger may be because many students lack new goals and are prone to boredom after starting university. Moreover, boredom is a hidden psychological state of suppression, which is an early sign of anger (Dahlen et al., 2004; Zhao et al., 2015). Therefore, undergraduates should make plans for college life and set reasonable goals to enrich their lives, avoid boredom, and reduce anger and aggressive behaviors.

The frequency of positive and negative emotion words had dynamic variations during a day, which synchronized with undergraduate students’ daily routines. The present study found that TT activity from 1 a.m. to 6 a.m. was low, and activity peaked at noon and about 10 p.m. This tendency is consistent with the results of Golder and Macy (2011), except that they found that Twitter activity peaked first at 9 a.m. rather than at noon. This inconsistency may be due to differences in the participants, as the participants of Golder and Macy’s (2011) research included not only undergraduates but many other groups as well. Most of them went to school or work and had limited time for social media after 9 a.m. Moreover, considering that the changes in individual emotions are closely related to their physiological rhythms and daily routines (Golder and Macy, 2011; Wang et al., 2016), the TTs reflecting undergraduate students’ life states can be important resources for analyzing their emotions.

There were intragroup differences in frequency and daily rhythms of emotion words. Grade was the main variable that affected the frequency of emotion words in TTs. As the grade increased, the frequency of emotion words in most subcategories in TTs decreased and the fluctuation in daily rhythms became smaller. We then performed an additional analysis about the user activity of Tencent QQ by students of all grades and found that both the average number of words and the average number of TTs per hour were the highest among freshmen, and the lowest among seniors and above. This result was consistent with previous research that senior students use social networks less frequently (Wang et al., 2015). However, the frequency of emotion words is an indicator excluding the influence of user activity of Tencent QQ. Therefore, the low frequency of emotion words in senior students’ TTs implied that senior students express less emotion on social networks. This result is similar to Lu et al. (2016), who found that students from higher grades express less emotion.

There was no significant difference in the frequency and daily rhythm of emotion words used in TTs based on having had a left-behind experience, which is inconsistent with existing research results (Li et al., 2009). The reason for this inconsistency may be that previous studies mostly used self-report scales to evaluate emotions, which is a retrospective evaluation induced by the outside world. In contrast, the data in this study came from TTs, which are spontaneous and real-time records that are not affected by recall bias. A study showed that the retrospective ratings of emotion are significantly higher compared to concurrent ratings for unpleasant conditions (Bruun and Ahm, 2015). Although there was no significant difference in emotion in daily life between undergraduates with or without a left-behind experience, it is possible that undergraduates with left-behind experience would have more retrospective emotional distress.

The present study explored the daily rhythms of emotions expressed by undergraduates in their Tencent tweets. To our knowledge, it is the first study to analyze word frequencies on a social network other than Twitter, Facebook, or micro-blog. Thus, this study has added new evidence that text on social media can be an important resource for analyzing undergraduates’ emotions. Adding to studies from individualistic cultures, the present study provides insights into emotion expression on social networks in a collectivist culture (China). The most important difference found in this respect is that Chinese individuals are more likely to express love on social media in comparison to real situations. In addition, the present study has also theoretical and practical implications.

The findings of the daily rhythms of positive and negative emotions showed that positive and negative emotions tended to have opposite trends from 12 a.m. to 2 a.m., 6 a.m. to 7 a.m., and 8 a.m. to 11 a.m. (i.e., as the frequency of one type of emotion word increased, the other decreased), and the same trend from 2 a.m. to 6 a.m., 7 a.m. to 8 a.m., and 2 p.m. to 9 p.m. (i.e., the frequencies of both positive and negative emotion words either increased or decreased). From 11 a.m. to 2 p.m. and 9 p.m. to 12 a.m., the trends were unrelated. These results are inconsistent with Russell’s (1980) circumplex model of affect, which considers that positive emotion and negative emotion are opposite poles of the valence dimension. According to the model, as one kind of emotion falls, the other rises. Our results support the positive and negative affect model proposed by Watson and Tellegen (1985) and the evaluation space model proposed by Cacioppo and Berntson (1994). According to Watson and Tellegen’s (1985) model, positive and negative emotions are two independent dimensions. Positive emotion is a combination of pleasure and high activation, while negative emotion is a combination of displeasure and high activation. According to the evaluation space model by Cacioppo and Berntson (1994), positive emotion and negative emotion are usually antagonistic to each other, but they may have uncoupled activation, co-activation, or co-inhibition. That is to say, a change in positive emotion may likely be the same as that in negative emotion, or it may occur in parallel or reverse with a change in negative emotion.

These notions provide directions for emotion regulation strategies. Previous researches focused on the strategies for decreasing negative emotion, such as reappraisal, suppression (Gross, 1998), and cognitive distraction (Strauss et al., 2016). However, our results suggested that the strategies of increasing positive emotion and controlling negative emotion are both essential when regulating emotions. Reappraisal, which was recently proved to be effective for decreasing negative emotion and increasing positive emotion, is a better strategy than distraction and suppression (Troy et al., 2019).

We found that the trough period for the average number of words in a single TT per hour was not from 1 a.m. to 6 a.m. but from 5 a.m. to 6 a.m. This means that the total number of words undergraduates published in a TT from 1 a.m. to 4 a.m. was low, but the average number of words per TT was high. The peak of the frequency of negative emotion words was from 1 a.m. to 4 a.m., and the trough of the frequency of positive emotion words was in the same period, which is consistent with previous research. Many studies have shown that late sleepers or those who have sleep disorders (i.e., individuals who are often awake from 1 a.m. to 4 a. m.) have high rates of depression, anxiety, and other emotional disorders (Alvaro et al., 2013; Zhang et al., 2017; Guo et al., 2018). Moreover, studies show that poor sleep quality can predict suicidal ideation (Supartini et al., 2016). Thus, it is reasonable to infer that undergraduate students who are active Tencent QQ users from 1 a.m. to 4 a.m. are likely to publish negative emotion words and are vulnerable to mental disorders. For students who are often active during the wee hours, more efforts should be made to adjust sleep rhythms and regulate emotions. It is also wise to seek help when necessary. The school administration department should also pay more attention to these students.

Our study is not without limitations. The first limitation was the representativeness of the sample as all the students were from a single university, and the number of students from the senior grade was less. The study should be replicated in the future using a more representative sample. Moreover, we did not explore the relationship between emotion expression on social networks and individuals’ mental health. Thus, it is still an open question whether emotion expression on social networks can predict mental health. Emotion expression may be a reflection of individuals’ current mental states, and emotion expression on social networks may be a strategy of emotion regulation that has positive impacts on individuals’ mental health. Further research should be conducted to determine the degree to which emotion expression on social networks is associated with mental health. Finally, text analysis based on word frequency in present study has been widely used and its validity was confirmed, but more text analysis methods have been developed, such as machine learning approach (Lee et al., 2019). These new methods should be alternative approaches to analyze the emotion in undergraduates’ texts on social media.

Tencent Tweets provide important material for analyzing the state and variation of undergraduate students’ emotions. Undergraduates tweeted more positive emotions than negative emotions on TTs. The emotions in undergraduate students’ tweets changed considerably during the wee hours but were fairly stable during the day. Positive emotions increased more intensely than negative emotions during the day as each hour progressed. There were also intragroup differences in the use of emotion words in TTs. The late sleepers who are likely to post negative emotions should be paid close attention. As positive and negative emotions were not polar of one dimension, according to our results, strategies that act on both positive and negative emotions should be adopted when regulating emotions.

The raw data supporting the conclusions of this article will be made available by the authors, without undue reservation.

Ethical review and approval was not required for the study on human participants in accordance with the local legislation and institutional requirements. The patients/participants provided their written informed consent to participate in this study.

Both authors designed the study, involved in data collection, and read and approved the final version of the manuscript. R-XL constructed the emotion dictionary for TT text. HL performed the statistical analysis. R-XL wrote the Introduction and Discussion sections of the first draft of the manuscript and HL wrote the remaining the sections.

The authors declare that the research was conducted in the absence of any commercial or financial relationships that could be construed as a potential conflict of interest.

All claims expressed in this article are solely those of the authors and do not necessarily represent those of their affiliated organizations, or those of the publisher, the editors and the reviewers. Any product that may be evaluated in this article, or claim that may be made by its manufacturer, is not guaranteed or endorsed by the publisher.