From Hashtag to Healing: Exploring Mental Health Support in Academic Twitter Networks

Abstract

This study explores the dynamics of #AcademicTwitter, investigating user connectivity and mental health attributes amidst pre-COVID-19 societal pressures and the subsequent pandemic onset. Analyzing key actors, we find that institutions and content-driven accounts are more prominent figures in shaping the discourse within this hashtag highlighting a larger role in community engagement. Language analysis reveals stable stress and wellbeing markers but fluctuating motivation markers, notably during pandemic-affected months. Central actors bridge information dissemination with addressing academia’s mental health needs with many using humor and satire to alleviate stress. Our findings suggest that academics used #AcademicTwitter’s online community to communicate, oftentimes with humor, and find mental health support through the pandemic and shortly thereafter.

Plain language summary

Analyzing tweets to see relationships between mental health and the COVID-19 pandemic within the academic Twitter social network

This study looks at how people interact on #AcademicTwitter, focusing on connections and mental health before and during the COVID-19 pandemic. We found that institutions and accounts focused on content play a big role in the conversations within this hashtag, showing their importance in community engagement. Analyzing the language used in tweets, we observed that stress and wellbeing levels stayed stable, but motivation levels changed, especially during the pandemic months. Key users helped share information and address mental health needs in academia, often using humor and satire to reduce stress. Our findings suggest that academics used the #AcademicTwitter community to communicate, share humor, and find mental health support during the pandemic and shortly after.

Social media, as a source of news, community, networking, and leisure, is a constant feature today with nearly 72% of American adults using social media in some capacity in 2021 (Pew Research Center, Social Media Fact Sheet, 2021). Platforms like Facebook, Instagram, Twitter, and LinkedIn enable individuals to connect across distances, facilitating the exchange of ideas, resources, and support. These platforms create spaces where like-minded users can engage in discussions on topics of shared interest, be it through forum-style posts, images, or short-form message (Adams et al., 2011; Wang et al., 2016). Twitter, in particular, stands out as a dynamic social network where users actively shape and influence conversations, driving the narrative within the communities they are a part of.

The size, frequency, distribution, and diversity of data generated by Twitter make it s a unique opportunity to analyze language associated with mental health needs and emotions in an organic setting. Tweets offer a snapshot of the user’s immediate thoughts through Twitter’s short text format. With over 6,000 tweets sent per second (Wang et al., 2016), the high velocity of data is advantageous since it is stimulated by external events oftentimes occurring simultaneously to the action of tweeting. Moreover, the open-ended nature of Twitter allows users to engage in conversation on their own terms, making it a rich source of data for analyzing trends in language and sentiment around specific topics.

Despite the growing body of research on social media, there has been little exploration into how academics use platforms like Twitter, particularly in relation to mental health and well-being. Previous research (American Psychological Association, 2020; Klar et al., 2020; Gomez-Vasquez & Romero-Hall, 2020; Wright, 2015) has primarily focused on how academics use Twitter for professional promotion, communication, and knowledge sharing within the community of professionals in the academic community which includes a mixture of academic accounts, professors, graduate students, and universities, as well as other users that use the hashtag #AcademicTwitter. Klar et al. (2020) observed that academics have embraced Twitter as a means to engage with audiences beyond traditional academic journals, using the platform to disseminate ideas and foster public engagement. In addition to being an online portfolio, it is a platform for research sharing, class support, virtual conference media, and personal communication (American Psychological Association, 2020). However, less attention has been given to how academics use Twitter to discuss mental health challenges, such as stress, workload, and overall wellbeing.

The significance of studying academics’ behavior on Twitter is multifaceted. As social media continues to evolve as a tool for professional engagement, it offers unique insights into how scholars build communities, share knowledge, and support each other across geographical boundaries. In recent years, academics have turned to platforms like Twitter to not only advance their professional work but also to express personal challenges, including issues related to mental health, stress, and work-life balance. By examining how academics use Twitter, we can uncover patterns in how they navigate the intersection of professional development and personal well-being. This is particularly important given the growing discourse around mental health in academia, where stress and isolation are common. Understanding these patterns can shed light on how social media serves as a support system, especially during times of crisis or high pressure, and offers opportunities for fostering healthier academic environments.

Creating an online professional network enables individuals to connect with peers facing similar experiences, regardless of location or time zone. While traditional professional development and learning communities have provided a foundation of pedagogical and academic growth within academia, the role of online communities in providing mental health support for academics remains underexplored. By adopting a macro-level perspective, this study examines the broader systemic impacts of #AcademicTwitter, this study adopts a macro-level perspective, focusing on how the platform fosters a global academic community and shapes the collective experience across diverse institutions and geographies. In doing so, this study aims to answering the following research questions:

1. What is the social network structure of #AcademicTwitter, and how does it change over time?

2. What are the characteristics of the most active members of the social network within #AcademicTwitter?

3. How have academics using #AcademicTwitter discussed mental health topics in an online space?

Conceptual Framework

Our inquiry is guided by social network theory, which analyzes the relationships between actors and their communication within the network (Borgatti & Ofem, 2010; Mitchell, 1974; Scott, 1988). The relationship between actors are not isolated occurrences but rather a web of communication paths (Borgatti & Ofem, 2010) that can directly, or indirectly, affect others within the network (Scott, 1988). On Twitter, retweets and mentions create interconnections between users, linking their interactions. By examining the web of relationships within a social network, we can focus on the actors and communication links that revolve around the #AcademicTwitter hashtag. Central actors in the network, such as universities or highly followed individuals, can exert significant influence over other actors by amplifying or shaping discourse, especially in relation to sensitive topics like mental health. By identifying these central actors, our analysis highlights how social network structures contribute to the spread and discussion of mental health attributes within the academic community.

In examining how #AcademicTwitter is used to discuss mental health topics, we leverage social network theory to understand the patterns and structures of communication that facilitate these conversations. Academics, particularly those under pressure due to their profession’s demands, may rely on these outlets for discussing mental health concerns such as stress, burnout, and well-being. These conversations are not confined to isolated individuals but instead take place within a community where central actors and key influencers have the ability to shape the conversation and disseminate supportive or motivational content, such as memes, gifs, and personal anecdotes. As these mental health discussions circulate through the network, they create a web of interrelated communications that can either provide validation and support or propagate stress and anxiety, depending on the narrative being shared.

Additionally, to better understand the role of Twitter in mental health awareness, the sensemaking framework provides an avenue to determine how people process and interpret information (Weick, 1995). When confronted with a phenomenon, individuals extracts clues from the phenomena and their environment, using these clues to ask questions and interpret their personal explanation (Mills, 2008; Weick et al., 2005). The social network of a Twitter community, particularly one like #AcademicTwitter, creates an environment in which community members use sensemaking skills to communicate their interpretations of phenomenon—such as their mental health challenges—both to themselves and to the broader public. External stimuli, such as societal pressures or professional struggles, and internal cues, such as personal emotions, converge in these discussions (Helms Mills et al., 2010). Through retweets and mentions, users make sense of their experiences and helps others do the same, generating a dynamic feedback loop within the network.

The #AcademicTwitter community thus represents not only a professional network but also a mental health resource where uses collectively make sense of their personal successes, struggles, and emotional well-being. The flow of information within this network—enabled by retweets, mentions, and ongoing conversations—fosters collective narrative about academic life and mental health. By studying this feedback process, we can better understand how individuals within the community perceive and react to mental health topics, and how they use the societal structure of the network to seek support, share experiences, and promote mental health awareness.

Twitter

Twitter, now known as X, emerged as one of the more influential and powerful social media platforms with 22% of adults using Twitter within the United States since it began in 2011 (Pew Research Center, Social Media Fact Sheet, 2021). As of September 2021, the platform reported 288 million daily active users globally, including 73 million users in the United States (Pew Research Center, Social Media Fact Sheet, 2021; Wang et al., 2016). With 500 million tweets sent daily in 33 different languages, Twitter facilitates vast and diverse communication, with 77% of users residing outside the U.S. (Wang et al., 2016). In the U.S., Twitter’s user demographics largely reflects the population’s gender and racial distribution, with 53% identifying as female and 60% White, 16% Black, and 11% Latino (Wang et al., 2016).

Twitter is categorized as a “microblog” due to its 280-character limit. As such, it is an event-driven platform where individuals and media outlets can share news and engage in real-time discussion (Murthy, 2012). The brevity of tweets does not diminish the platform’s capacity for meaningful discourse; users often embed external links or use hashtags to connect with broader conversations. Hashtags, such as #AcademicTwitter, help users participate in professional communities, facilitative information exchange within a shared interest cluster (Gomez-Vasquez & Romero-Hall, 2020).

Twitter’s open, public nature allows users to broadcast messages widely, with retweets and hashtags enabling content to reach broader audiences without explicit consent from the original poster (Murthy, 2012). This creates a dynamic environment where users can interact with others beyond their immediate social circle, fostering connections based on shared interests rather than personal connections (D. M. Boyd & Ellison, 2007).

Despite its character limit, Twitter users engage in deep conversations through these networks, with users gaining visibility, followers, and affirmation through retweets and likes (Shepherd et al., 2015). For academics, the platform serves as a tool for collaboration, reflection, and engagement across disciplinary boundaries (Malik et al., 2019), helping them connect with peers and participate in discussions that may otherwise remain siloed within traditional institutional structures.

Twitter and Health Research

Perhaps surprisingly, Twitter is a suitable data source for health research, particularly the realm of mental health, as users often openly share personal thoughts and experiences (S. Paul et al., 2011; M. J. Paul & Dredz, 2011; Wang et al., 2016). This openness stems from Twitter’s wide audience and the perceived accessibility of experts and peers within various fields (Park et al., 2016; S. Paul et al., 2011). Through hashtags, direct messages, or mentions, users can reach out to others, including professionals, creating a line of communication that makes health discussions more public and collaborative. This public nature of Twitter thus facilitates real-time interactions, making it a valuable tool for collective data on how individuals express mental health concerns and access support through social networks.

Research has shown that approximately 10% of Tweets contain statements or questions related to health and wellness (M. J. Paul & Dredz, 2011). This makes Twitter a significant repository for tracking public health issues, with studies using the platform to monitor flu outbreaks (Culotta, 2010; M. J. Paul & Dredz, 2011), sports concussion protocol (Sullivan et al., 2012), and tobacco use (Prier et al., 2011). Furthermore, language analysis has been used to predict life satisfaction through keyword analysis (Schwartz et al., 2013) and to identify work stress and emotional states (Wang et al., 2016). These examples underscore Twitter’s potential to capture and analyze health-related conversations in a large, diverse population.

However, the use of Twitter for mental health research does raise important concerns about privacy, the authenticity of information shared, and the sensitivity of health-related topics. While many users are willing to share personal experiences, the public nature of the platform means that privacy is not always guaranteed, and individuals may withhold or alter information due to concerns about being judged or misunderstood. Additionally, the authenticity of the information shared may be compromised as users may craft their tweets to align with socially desirable narratives rather than genuine experiences.

Academia Mental Health

In the context of this study, the above concerns are particularly relevant as we examine how academics use #AcademicTwitter to discuss mental health topics. Especially since mental health is a widespread concern throughout the profession. 70% of the US workforce reports that their job includes high levels of stress (Clay, 2011) Health consequences of workforce related stress are concerning with heart disease as the leading cause of death for most age, gender, and racial groups in the United States (Centers for Disease Control and Prevention [CDC], 2018). Work stress contributes to other physical and emotion problems such as headache, anxiety, insomnia, depression, burnout, and counterproductive behaviors (Spector & Fox, 2005; Wang et al., 2016).

Academic professionals experience high levels of stress perhaps related to the “publish or perish” mindset which may have a negative effect on mental health (Miller et al., 2011). Low job satisfaction and burnout in academia has been related to emotional exhaustion and reduced personal accomplishment (Kumar, 2015). Additionally, mental detachment from the job, feelings of cynicism, and low self-worth hinder feelings of personal accomplishment (Maslach et al., 2001) influencing low productivity and high stress. Perceived burnout can be correlated to lower publication numbers (Kumar, 2015; Takahashi & Takahashi, 2010) which impacts mental health.

Performing a wide range of duties each day ranging from teaching, advising, grading, delivering course materials, to preparing for lectures, those in academia face challenges with time management to perform required duties. In addition to expectations as a teacher, there are research obligations such as grant writing, data management, writing, and publication. Coping effectively with stress can be helped with increased social support and interpersonal interactions (Cohen & McKay, 2020). Research suggests that burnout and stress levels are related to age and experience with academics over 45 years old experiencing lower levels than those new to the profession (Kumar, 2015). Indeed, younger academics, who may be more likely to engage in social media, might benefit most from the #AcademicTwitter community for mental health needs.

Attempting to give a questionnaire or survey to academics about their mental health may result in biased results; however, big data directories such as Twitter allow for a robust data source that is unobtrusive. Twitter provides real-time reports of individual thoughts and feelings, particularly about stressors, mental health needs, and celebrations. Though public in nature, research through tweet analysis regarding mental health may reduce bias giving insight into the profession, particularly regarding impacts of the COVID-19 pandemic era.

#AcademicTwitter

While many forms of social media involve linking profiles of individuals with a common personal interest or relationship, Twitter users connect through using hashtags to reach a broader audience creating a network with a common theme. A key aspect of Twitter conversations is the formation of online community where users sharing information and foster relationships with others of similar interests (Xu et al., 2015). Academic driven hashtags, such as #AcademicTwitter, #AcademicChatter, #PhDvoice, and #PhDchat, have become increasingly popular among academics and students. These large support networks allow users to engage in discussions about shared experiences, creating spaces for discourse around academic themes (Shepherd et al., 2015). Research shows that about 1 in 40 academic scholars use Twitter for collaboration and community building (Wright, 2015). Gomez-Vasquez and Romero-Hall (2020) found that micro-communities within #AcademicTwitter are often centralized, with few participants driving most conversations. Approximately 40% of hashtag users tweet more than once with top contributes including educators, media platforms, and other professionals (Gomez-Vasquez & Romero-Hall, 2020) which fosters an online global professional development network across educational levels (Malik et al., 2019). Twitter serves as a “virtual water cooler,” providing a space for scholarly exchange, networking, and community building (Wright, 2015).

Themes such as accessibility, academic life experiences, and teaching support frequently emerge within top contributor conversations (Gomez-Vasquez & Romero-Hall, 2020). Social media users through shared experiences, helping reduce stigma around struggles in academic life (Shepherd et al., 2015). Building on previous general descriptive research regarding the Twitter academic community (American Psychological Association, 2020; Klar et al., 2020; Gomez-Vasquez & Romero-Hall, 2020; Wright, 2015), this study explores how #AcademicTwitter functions as a community where academics communicate their mental health needs and concerns, examining the role of this network in addressing those challenges by focusing on language analysis metrics.

Methods

Data Source

Using publicly accessible data from Twitter, this study employed a Twitter data scraping technique using Twitter’s Application Programming Interface (API) through an Academic Developer License Profile obtained by the researchers. Scraping techniques involve creating an automated process of retrieving data quickly, and compiling and consolidating data into a structured output. The internet, and information contained on social media platforms, continues to grow exponentially. Obtaining information appropriately through Twitter’s API format is a formidable task due to the vast amounts of data that can be harvested (Baskaran & Ramanujam, 2018); however, creating a bot using Python Programing Language is an effective research tool when looking at how people interact within a social media environment. The twitter scraping bot used in this studied was created using Python to gather monthly tweet data during the research period.

Tweets gathered on Twitter using the search query “#AcademicTwitter” spanned a date range from January 2018 to December 2021. Data scraped within this query included the full tweet with hashtag mention, timestamp, tweet location, tweet ID, username, username ID, and numbers of retweets, tweet likes, and replies. This data was exported into two .CSV files per scrape and merged. For the purposes of our research question, we were interested in querying tweets the most recent tweet depository and within a range surrounding the COVID-19 pandemic; therefore, our tweet query began approximately 1 year prior to the COVID-19 pandemic shutdowns and continued until December 31, 2021 to determine if #AcademicTwitter was a social network community impacted by the pandemic.

The number of users identified in this query, including those that tweeted and those that were mentioned within a tweet, was 2,703,896 users total. For the purposes of this study, duplicate users were omitted for the node list in our network graph analysis. After omitting duplicate users, our data included n = 528,496 individual users as identified by User ID within the Twitter API scrape. Omitting duplicate users did not limit our data. All users within the social network were maintained. Omitting duplicates within the user list allowed connections between users to receive a ranking which illustrated their activity within the network. This means that 44.5% of users used #AcademicTwitter more than once in this data

For our initial analysis of tweets, we only included tweets and user information from single tweets and omitted retweets and mentions within the sample which led to an omission of 1,284,186 tweets using the #AcademicTwitter community as a retweet. Therefore, our initial study sample included a total of N = 642,764 original tweets. For this total number, Figure 1 shows the total number of tweets per month in this sample. This illustrates the gradual use of #AcademicTwitter in 2018 with a sharp increase in February 2019 along with additional increases and decreases in usage until the next sharp increase in March 2020. Additionally, another sharp increase occurred in March 2021 followed be a decline in usage. The data set containing only tweets was used to analyze language use and psychometric constructs of tweet text in this social network specific to answering RQ3.

Figure 1.

Number of original tweets each month.

To correctly portray the entire social network of #AcademicTwitter, retweets and mentions were also scraped using the same Twitter API algorithm and similar bot written using Python. When analyzing the full query (tweets and retweets) the sample size increases to N = 1,926,950. Figure 2 shows the number of #AcademicTwitter uses per month during the query range. Figure 3 compares the values of original tweets and the tweet/retweet total gathered. This complete query was used to analyze the social network created during this period.

Figure 2.

Number of original tweets and retweets each month.

Figure 3.

Comparison of tweets and retweets.

Methodology

This study employs a mixed-methods approach to analyze the role of #AcademicTwitter in both community building and mental health discourse within academia. We use social network analysis (SNA) to examine the structure of interactions among users and text classification to investigate the language used in tweets, particularly around mental health topics. The methodology draws upon the analysis plan developed by the #CommonCore project (Supovitz et al., 2017), which similarly used SNA and content analysis to examine networked public discourse around the implementation of Common Core.

First, we used Stata for preliminary statistical analysis and Gephi (Bastian et al., 2009) to visualize and analyze the social network structure formed by users of the #AcademicTwitter hashtag. and to identify measures of centrality and closeness of ties within the network, as well as key actors in the network. This involved mapping connections between users to identify measures of centrality (which determines influential users) and closeness of ties (which evaluates how connected individual users are to others in the network). This analysis allowed us to identify key actors in the network, such as universities, content-driven accounts, and individual users, based on their degree of influence within the network. Through Gephi, we could examine patterns of retweeting, mentions, and direct replies, revealing both the highly interconnected nature of the community and the central figures driving discussions.

Second, we use Linguistic Inquiry and Word Count (LIWC; Pennebaker et al., 2015) to analyze the textual content of the tweets, focusing on the frequency and context of words related to mental health, such as stress, motivation, and well-being. LIWC allowed for a systematic exploration of how academics express their emotional and psychological states in their tweets. By tracking changes in the use of mental health-related terms over time, we were able to assess how external pressures, such as the COVID-19 pandemic, influenced the overall tone and mental health discourse within the community.

These combined methods provide a macro-level understanding of the network’s structure and a micro-level analysis of the content and sentiment within the community, offering a comprehensive view of how #AcademicTwitter functions as both a professional network and a space for addressing mental health concerns. The integration of SNA and LIWC ensured that we could not only identify the influential actors within the community but also gain insights into the mental health needs and concerns expressed by individual users. Together, these methods enabled a robust analysis of the interactions and content that define the community.

Graph Structure

To visually understand a social network, a network graph was created using Gephi, an open-source program specific to network data, to illustrate the relationships between members of a specific network. In a network graph, the relationship between members is represented by a node (user) and a line/edge connection. In this study, our line/edge is a tweet or retweet. As data is added in a network graph over time, the number of nodes and edges increases as the number of relationships within the network increases. This creates clusters of users due to their smaller networks organically formed within the larger social network community that is “unmistakable” and a “pattern in social media toward clusters into insular like-minded communities” (Lynch et al., 2014; Willis et al., 2015).

The network graph is constructed using a node and edge list allowing us to see who is interacting with whom. The node list consisted of unique usernames, corresponding user Id, and user location. Duplicate user observations were removed using Stata and the .DTA file was exported as a .CSV file for use in Gephi. Another .CSV file was exported using Stata which included all original tweets. The tweet file was used as the edge list in Gephi and contained the user Id for the original tweet and mention usernames in a directed graph format. Since tweets and retweets are directional, a directed graph format was appropriate for this data set where nodes represent users and edges represent retweets and mentions.

Each interaction is represented in the network graph as a line connecting two nodes. Each node represents a user, and each line represents a connection between the users. For example, in a directional graph this means that user (node) A has a line (edge) directed toward user (node) B which signifies that node A shared a connection with node B. This connection could be a reply, retweet, or mention.

Since a twitter user can tweet, retweet, and be retweeted many times, it was important for each node to be unique to a single user and duplicates removed from the user list containing all tweets harvested. Because each node represents a unique user, multiple edges can begin from a single node which illustrates how one user can tweet, retweet, and be retweeted within the same social network infinite times. Therefore, our total social network analyzed includes 528,496 nodes corresponding to unique users within the data set and 1,926,950 edges representing tweets, retweets, and mentions.

A visual representation of the network is found in Figure 4. Those users found in social network clusters are shown in Figure 4 with different colors to highlight their connections. Edge thickness can also be seen in Figure 4 which illustrates the number of times that User A tweeted to User B. Most edges are thin which represents only a single tweet. This network graph was created by ranking nodes by degree rank creating larger nodes for those more active users.

Figure 4.

Social network graph of #AcademicTwitter.

Network Metrics

Modularity

After a network graph is tabulated, a statistical analysis of the modularity of the graph can be performed to determine if the network graph is one large network in which all nodes equally interact with one another or if clusters of smaller communities form within the larger network (Willis et al., 2015). These clusters of sub-community interactions represent closer relationships within the larger conversation with users more interconnected by tweet replies, mentions, or retweets (Blondel et al., 2008).

Modularity calculations determine to what extent a large network can be decomposed into sub-networks (Blondel et al., 2008; Willis et al., 2015). The modularity of #AcademicTwitter resulted in the identification of 54,825 distinct modular communities; however, due to the size of the network, the largest 15 communities are highlighted with color in the network graph. Additionally, the modularity score of #AcademicTwitter social network community was 0.639 which is within modularity limits and refers to the sum of edges inside a cluster compared to the expected edges thought to exist as determined by if the network were a random assortment with the same number of nodes and edges (Newman, 2006). The largest community was formed by 10.63% of all nodes (Figure 5).

Figure 5.

The top 15 modular communities ranked by their percentage of the entire network.

By using modularity as a network parameter, we can see that only one of the sub communities contributed a significant amount of the social network with 10.63% of the total network cluster or 56,179 nodes. The remaining subcommunities in #AcademicTwitter were smaller with the next community including 5.56% of the total network or 29,384 nodes.

Centrality

Measures of centrality allow us to see how significant a “node” is within the social network which helps us understand the central actors using #AcademicTwitter in this period. Network graphs represent the heterogeneous structure of social networks with differing roles played by unique actors within the network (Gómez, 2019). Those nodes with the highest centrality values represent users with the highest degree of influence within the network (Gómez, 2019; Valente et al., 2008).

While there are many ways to calculate centrality (Barabasi & Albert, 1999; Valente et al., 2008; Willis et al., 2015), for the purposes of this study we will use degree centrality, eigenvector centrality, and page rank to analyze the networks which allows us to understand the influence of specific accounts and/or clusters within the larger social network community.

Degree Centrality

To determine how many connections are made at a specific node, degree centrality is calculated to understand relationships between nodes (Barabasi & Albert, 1999; Freeman et al., 1979). Perhaps the most simplistic measure of centrality, degree centrality calculates how many direct connections each node within the network contains (Gómez, 2019). Degree centrality is appropriate for undirected network graphs; however, additional centrality checks provide more details for directed networks (Barabasi & Albert, 1999).

Betweenness Centrality

One way to determine the potential of an actor to influence others within the network is to calculate betweenness centrality which refers to how often a node is found between any two nodes within the network (Anthonisse, 1971; Freeman, 1977; Willis et al., 2015). An emphasis is placed on nodes which falls when the shortest paths within the network are examined (Gómez, 2019) which could project the ability of the actor to control the flow of information within the network. The higher betweenness centrality of a node within this path may indicate that this actor has a higher potential to influence those actors surrounding within the social network (Freeman, 1978; Newman, 2006; Willis et al., 2015).

Eigenvector Centrality

Eigenvector centrality is another measure of a node’s importance within the network graph and the relative influence that a particular node has on the network (Boccaletti et al., 2014). Not only does eigenvector centrality consider the topological position of the node within the network, but it also calculates the relative influence that nodes within the cluster have on the network graph as well (Bonacich, 1972; Gómez, 2019). A high eigenvector centrality means that a node is pointed to by many other nodes and connected by more edges that other nodes within the network (Boccaletti et al., 2014). Gephi is used to run this statistical analysis which allows those nodes with a higher eigenvector centrality to be denoted by a larger diameter node than those with a smaller eigenvector centrality (Bastian et al., 2009).

PageRank Centrality

PageRank and eigenvector centrality indicate similar values. PageRank uses the indegree centrality as the main measure to estimate the influence of a node on the network graph, especially in a directed network graph (Brin & Page, 1998). It is a specific measure of eigenvector centrality; however, PageRank considers the source of the edge giving more importance to those edges received by influential nodes within the network graph (Gómez, 2019).

#AcademicTwitter Data

For the #AcademicTwitter social network, the average degree is 1.808 which refers to the average number of edges connected to a node throughout the network (Wasserman & Faust, 1994). Additionally, the average weighted degree calculated is 2.532 which refers to the average sum of the weights of edges connected to a unique node (Wasserman & Faust, 1994). The greater the weight of an edge corresponds to more interactions between two nodes. Using centrality, we determined central actors in the #AcademicTwitter social network.

Those accounts that were most active users within this study spanned a range of accounts, including businesses, research journals, higher education institutions, professors, and graduate students (Figure 6). Additionally, the top 5 active accounts using #AcademicTwitter are content accounts aimed at increasing community for academics by academic support (@PhDVoice, @AcademicChatter, and @openacademics) and humor through gifs and memes (@thephdstory and @Dr_Meming). 17 users of the top 50 actors are individual use accounts. The top 5 actors within the network created approximately 2.7% (52,349) tweets within the data. Increasing to the top 50 actors within the network accounts for 7.6% (145,757) tweets.

Figure 6.

Thirty most active users (tweet or retweet) within social network.

The top ranked accounts by degree centrality contain several of these actors which were identified as the most active users within the social network (Figure 7). Since degree centrality refers to the number of direct connections to the network, it makes sense that there is overlap between the most active users of the hashtag and those with the greatest number of connections within the community.

Figure 7.

Highest 15 Degree Centrality values by username.

Another measure of influence within a social network, eigenvector centrality, identifies nodes with high levels of edge connections. A higher eigenvector centrality measurement signifies that the user has more connections (retweets and mentions) within the community. As seen in Figure 8, this corresponds with Degree Centrality leading us to understand that those with the most direct connections also have the highest levels of edge connections.

Figure 8.

Highest 15 Eigenvector Centrality values by username.

Finally, PageRank can be considered a significant measure of page importance. Based on the links made to the node, this measure looks at the in-degrees (backlinks) for the node (Figure 9). Values of PageRank add to 1 in a social network with the higher PageRank scores signifying the more influential nodes within the social network. Though some user accounts are similar through all three measures of centrality, PageRank provides a different prospective since the number of links from other users creates a higher PageRank value.

Figure 9.

Highest 15 PageRank values by username.

Content Analysis Using LIWC

Content analysis is an appropriate approach for our study because we needed to investigate the content of tweet language to identify mental health related themes associated with our hashtag search. To convert tweet text into a numeric value to better understand constructs of interests, content analysis using the Linguistic Inquiry and Word Count (LIWC; Pennebaker et al., 2001, 2007) was used since it easily converts text to psychological constructs. The basic LIWC text analysis is counting words and calculating word frequencies. The following formula is used to calculate word frequencies in text:

$$\mathrm{LIWC}\mathrm{word}\mathrm{frequency}=\frac{\mathrm{word}\mathrm{counts}\mathrm{against}\mathrm{a}\mathrm{dictionary}}{\mathrm{total}\mathrm{word}\mathrm{count}\mathrm{in}\mathrm{a}\mathrm{text}}\mathrm{X}100\mathrm{\%}$$

where a dictionary is a collection of words, word stems, or phrases predetermined to measure a particular psychological construct within the research question. Using LIWC’s built in dictionary, there are different component measures such as tone, positive words, negative words, affective processes, personal concerns, and social relationships. Additionally, purposeful dictionaries can be created or uploaded for specific studies. This iterative coding method was used to better understand themes and patterns within tweets.

For the purposes of this study, we first used the built-in LIWC dictionary for basic analysis of the #AcademicTwitter tweet file (Pennebaker et al., 2001). Second, we used a stress dictionary created by Wang et al. (2016) to analyze stress related words and negative emotion constructs associated with stress levels. Third, we used a well-being dictionary created by Ratner et al. (2021) to analyze linguistics associated with perceived well-being and positive and negative emotion constructs associated with mental health.

Emotionality

Positive and negative emotion words help us understand how people experience external and internal stimulus. Emotional response can vary radically from person to person, but nonetheless is a measurable linguistic measure synthesizing a person’s thoughts and feelings about an event or idea. LIWC is able to identify emotion through text by compartmentalizing some words as positive (e.g., nice, happy, love) and others as negative (e.g., sad, mean, hurt) when writing about an experience (Tausczik & Pennebaker, 2010). Figure 10 shows positive and negative text emotions within the tweet text file.

Figure 10.

Level of stress word indicators per month.

Pronoun Usage

Pronouns used within text can signify different psychological correlations in LIWC. First person singular correlates to depression, emotional, and honest text; first person plural correlates to social group attachment; second and third person correlates social interests, awareness, and supports (Tausczik & Pennebaker, 2010).

Stress Level

Tweet files were examined using the stress dictionary created by Wang et al. (2016). This dictionary identified words associated with high levels of stress as an indicator that this health attribute might be shared within a social media network. Figure 10 shows stress indicated usage within the tweet text file. This dictionary also examines the use of question marks and exclamation marks within a text file to determine if there are questions being asked or achievement exclaimed. Though exclamation marks are used more that question marks within the tweet file source, there is no significant usage of either punctuation type.

Wellbeing

A dictionary created by Ratner et al. (2021) used indicator words to assess an individual’s perceived wellbeing in text as well as their motivation as indicators of mental health. Tweet files were examined on each of these indicators. Figure 11 shows perceived wellbeing of users and Figure 12 shows motivation indicators of users.

Figure 11.

Levels of wellbeing word indicators per month.

Figure 12.

Levels of motivation word indicators per month.

Findings

Big data analysis often reveals insights into the social network structure and key actors within #AcademicTwitter, answering the first two research questions. Between 2018 and 2021, #AcademicTwitter was used in 1,926,950 tweets creating a diverse community of academics, including professors, graduate students, universities, research journals, and informational content creators. The retweet rate within this community is significantly higher at 66.6% compared to the general Twitter retweet rate of 1.44% (Zarrella, 2009). This high retweet rate suggests that the information tagged with #AcademicTwitter is demed relevant and important by its community, indicating a strong engagement around academic topics.

Our analysis also looked at the use of the “@”-sign to address tweets to specific users, with approximately 75.$% of all tweets in the sample containing the “@”-sign. This high percentage reinforces the interactive nature of the #AcademicTwitter community which aligns with previous findings that the “@”-sign facilitates discussions within the network (Honey & Herring, 2009; Tumasjan et al., 2010) When retweets are omitted, 10.6% (205,075) of original tweets contain an “@”-sign suggesting that even outside of retweets, directional conversations are an important aspect of the network’s interactions.

The centrality analysis of top users reveals that while 11 of the top 15 users were content focused accounts (tweeting informational content, memes, and gifs), the most influential nodes in the network, as indicated by higher PageRank and eigenvector centrality values, were individual users. In fact, 10 of the top 15 values for PageRank and eigenvector centrality were from individual users rather than informational usernames. The suggests that personal accounts, rather than just content providers, play a significant role in shaping conversations within the community. For example, @OpenAcademics, a community account dedicated to supporting academics, ranked highest in all centrality measures and had a follower county of over 90,000.

@OpenAcademics describes the account as a “Twitter community to support fellow academics across all disciplines” including #MentalHealth (Open Academics, n.d.). Content analysis of the Twitter profile reveals a mix of information tweeted including writing resources, retweets from followers, memes and gifs, and celebrations of achievement in academia (Open Academics, n.d.). Similarly, @AcademicChatter, the top user within this social network, content is driven by retweets of followers. Additionally, another top ranking account in both tweet numbers and centrality analysis, @Dr_Meming’s content is almost exclusively memes and gifs about academia.

Our content analysis using LIWC gives insight into how academics have expressed mental health needs related to stress, workload, and motivation. Although the volume of tweets and retweets increased during the pandemic, LIWC analysis suggests that stress and wellbeing word markers remain at a constant level throughout the sample period. However, a significant decline in motivation word markers was observed beginning March 2020 until January 2021 which aligns with heightened societal pressures associated with the COVID-19 pandemic. This pattern indicates a possible emotional toll within the academic community during this time.

Interestingly, the most active accounts were those tweeting memes and gifs about academic life, while individual users were more likely to engage through retweets and mentions. This suggests that humor, often expressed through memes and gifs, was a central coping mechanism for academics, especially during the pandemic. Despite a relatively stable use of stress and wellbeing language, the constancy of these markers might reflect an ongoing struggle, while the use of humor and sarcasm indicates a complex, perhaps understated, expression of mental health needs.

Social network theory posits that networks consist of individuals with shared interests, and our findings align with this by showing that #AcademicTwitter participants likely used humor and retweeting as a form of sensemaking during the pandemic. The use of memes, gifs, and retweets as comping mechanisms suggest that academcis turned to informal, lighthearted content as a way to alleviate the pressures of their professional lives. This aligns with research indicating that humor can serve as a coping strategy during stressful situations, helping individuals process negative emotions. However, the decrease in motivation word markers alongside the constancy of stress and wellbeing markers points to a concerning undercurrent: while humor may have provided temporary relief, it likely masked deeper, unresolved mental health challenges. This raises important questions about whether #AcademicTwitter, while valuable for creating connection and community, adequately supports meaningful discussions around mental health or whether it inadvertently encourages surface-level engagement through human and sarcasm.

Discussion

Our study of the social network community #AcademicTwitter has attempted to understand not only the users and connectivity of contributors to the conversation on Twitter but also to identify mental health attributes of the users. Specifically focusing on the months prior to the onset of the COVID-19 pandemic, the months of increased societal pressure with government shutdowns, and subsequent months following the reopening phase, allows us to illustrate how the #AcademicTwitter community responded to adversity in academia.

The key actor analysis suggests that many of the key actors within this social network were not individual user accounts but rather universities, research journals, and content accounts that attempted to connect academics through the propagation of information, memes, gifs, and retweeting follower content. This suggests that content accounts can shape the network by disseminating information and follower retweets to further discussions within the community. The centrality of key actor accounts indicates that some accounts increase connectivity through dissemination of information and follower retweets, while others use a more individualistic approach to content creation.

Language analysis of mental health constructs using publicly available big data from Twitter provides organic insight into the perceived mental health needs of academia. Analyzing tweets through stress, well-being, and motivation dictionaries suggested that stress and well-being word markers did not fluctuate during the sample period, while motivation word markers did fluctuate, especially during the 10 months most impacted by societal pressures related to the pandemic. Central actor accounts, in particular, played a key role in responding to these needs by sharing memes, gifs, and satirical content aimed at providing comfort through humor to this to this community. This underscores the significance of intrapersonal relationships and informal peer support within a high-stress occupation.

The findings of this study are significant in understanding how online academic communities, such as #AcademicTwitter, operate as both professional and emotional support networks. The fluctuating use of motivational language in response to external societal pressures, such as the pandemic, highlights the critical role that digital platforms play in supporting mental health during periods of crisis. Furthermore, the prominence of universities and content-driven accounts as central actors within the network reflects how institutional presence can shape discourse and provide continuity during times of professional and personal upheaval. This research contributes to the broader understanding of how social media can serve as a vital support system for academics, offering insights that extend beyond formal academic settings to include emotional well-being and personal connection. The results suggest that social media platforms, like Twitter, can fill a gap in addressing the mental health challenges of scholars, offering a space for real-time, community-driven interaction and support.

In light of these findings, it becomes clear that while #AcademicTwitter serves as an important outlet for academics to connect and cope with the pressures of their profession, there are limitations to the depth and authenticity of the mental health discussion taking place. The reliance on humor and the performative aspects of social media may create a barrier to more vulnerable and honest conversations about the mental health struggles many face. As such, this study highlights the need for more targeted mental health supports within academic networks—both online and offline—that encourages deeper engagement with these issues and creates a culture where seeking help is normalized rather than stigmatized.

Implications

The findings of this study have several important implications for both academic institutions and individual scholars. First, the role of #AcademicTwitter as an informal support network highlights the need for institutions to recognize the potential of integrating digital platforms as part of their mental health support strategies. Given that many academics turn to Twitter, LinkedIn, and other social media platforms for peer support, humor, and motivation during times of stress, institutions could benefit from fostering similar networks internally or collaborating with existing online communities to provide more formalized mental health resources.

Additionally, the prominence of content accounts and institutional actors in driving discourse on Twitter suggests that universities and journals have the potential to play a more active role in supporting their academic communities by promoting not only professional content but also content related to well-being and emotional support. This study also suggests that future research should focus on how the use of social media by academics during crisis can influence long-term mental health outcomes and professional satisfaction. As digital engagement continues to grow, both individual scholars and institutions must consider how these platforms shape academic culture, networking, and emotional well-bring.

Limitations

Twitter analysis, particularly in the realm of educational research, is a new approach to understanding academic communities. Though it provides access to a large repository of big data, several limitations exist. First, Twitter analysis is restricted to individuals who have Twitter accounts. Wright (2015) determined that only 1 in 40 academic scholars use Twitter for collaboration, meaning the platform does not capture the full range of academic discourse. Furthermore, with 60% of Twitter users under the age of 35 (Wang et al., 2016), younger, tech-savvy scholars may be overrepresented, while older, potentially less tech-savvy scholars might be underrepresented, which may skew the dataset.

Second, Twitter posts reflect only the content users are willing to make public. The public nature of the platform incentivizes users to crazy their tweet to garner likes, retweets, and engagement, which could lead to the dissemination of socially desirable or curated content rather than genuine scholarly discourse. This can introduce falsehoods, omissions, or exaggerations, as users may present an idealized version of their professional lives, thought, or mental state. Such narratives might not always reflect the complexities or challenges faced by scholars in private.

Third, the study takes place during an extraordinary period marked by the COVID-19 pandemic, a time when academic discourse, professional practices, and personalized challenges were significantly impacted. It is important to acknowledge that the data may capture an unusual period of heightened online activity, particularly as many academics shifted their professional interactions to digital platforms. As such, the findings may reflect this unique period rather than the typical use of Twitter within academic circles.

Finally, the unstructured and dynamic nature of Twitter posts further complicates the analysis. Tweets are limited in length and often lack the depth of traditional scholarly discourse. The chaotic and spontaneous nature of the platform may result in fragmented or incomplete narratives, potentially leading to a biased analysis of academic trends and interactions. These limitations highlight the need to approach Twitter-based research with caution, acknowledging the potential for both skewed representation and incomplete data. Nevertheless, Twitter offers a unique, real-time glimpse into the interactions and concerns of scholars across disciplines, providing access to conversations that might otherwise remain undocumented. As such, it serves as a valuable tool for capturing emerging trends and sentiments within academic communities.