How Tweets Reflect Emotional States

A Q&A with a UCIPT Postdoctoral Researcher

What do tweets typically look like when someone is sleep deprived?

We have not studied this topic before. Sleep quality or deprivation are usually captured using “direct” measurements such as biometric sensors (e.g., Basis sleep bands) or sleep-related questionnaires. Nevertheless, I think social media data (e.g., Twitter) may provide useful information about a person’s sleep quality. An obvious way to tell if someone is feeling sleep deprived is to check whether the person has tweeted about it. However, a major limitation of this method is that not everyone tweets about their sleep or how tired they feel. Therefore, the alternative method is to examine the time signature of a person’s Twitter use. In one of our recent studies, we found that individuals who tweeted between 2:00 A.M. to 6:00 A.M. had a lower sleep quality score than individuals who did not tweet during this time. Results from this study suggest that we will need to look beyond the content of the tweets.

What do tweets typically look like when someone is afraid?

There have been several recent studies published examining ways to extract emotions from tweets. Based on Paul Ekman’s work, there are six basic human emotions (i.e., anger, disgust, fear, joy, sadness and surprise). Being afraid can be categorized as fear. In a recent study, the classifier used to identify fear tweets can be accomplished by using unigram (e.g., specific key words) and topic modeling (e.g., tweets related to specific topics). The researchers found that the linguistic style of fear tweets evokes negative sentiment, words about political places, or indications of goals of obtaining power. It is important to note that people use different language and linguistic styles on Twitter. Future studies need to examine whether this model used to identify fear can be applied to different types of users who vary in age, education, and ethnicity.

What do tweets typically look like when someone is well rested?

If someone is well rested, it is likely that their tweets will be positive, outgoing, and engaging. Thus, I think their tweets may have positive sentiment and emotions (e.g., joy, happiness, love). These users may also tweet during regular working hours, produce original content, and engage with other users. Future studies would be needed to confirm these hypotheses.

What do tweets typically look like when someone is depressed?

Evaluating content is one of the methods used to analyze if someone is depressed. For example, a previous study found that increased negative sentiment and greater expression of religious involvement may suggest someone is depressed. Some users may even tweet about their depression-related thoughts (e.g., thoughts of hurting themselves). The second method to detect depression is to evaluate social media usage. One study found that a sudden decrease in social activity may indicate when someone is depressed. Another recent study reported participants who use social media very frequently (>121 minutes per day) have 2.7 times the likelihood of depression compared with users who view social media less (0–30 minutes per day). Overall, these studies help us understand the types of patterns that researchers should be looking for when identifying depression using social media data.

What do tweets typically look like when someone is preoccupied or busy?

This is a very interesting question. There are several hashtags (e.g., #busy, #productive) that may give an indication of how busy people feel. I think the feeling of being preoccupied or busy is extremely personal. Some people may feel positive, excited, or productive when they are busy, while other people may feel sad, stressed, or angry. Furthermore, if someone is busy, then that person may not have the time to tweet about it. Therefore, I think Twitter data may be better suited to capture people’s emotions as a result of their workload rather than the workload itself.

Leave a comment

Your email address will not be published. Required fields are marked *