Introduction

After the outbreak of coronavirus disease (COVID-19), the World Health Organization and most governments carried out official control measures, such as restricting population mobility, maintaining social distance, closing schools, suspending all non-essential production and commercial activities, and isolation. During the COVID-19 pandemic, these measures had a certain effect on preventing the spread of the disease, but also inevitably caused some mental health problems (Kowal et al., 2021; Matias et al. 2020; Riva and Wiederhold, 2020). The public faces multiple pressures related to health, daily life, and other aspects (Riva and Wiederhold, 2020; Karyotakis, 2024).

Public mental health problems during the COVID-19 pandemic, including those resulting from prevention and control measures, have attracted the attention of many international organizations, such as the World Health Organization (Ghebreyesus, 2020). Therefore, some scholars proposed that virtual reality (VR) technology, which is not limited by location, can help people cope with mental health problems during the COVID-19 pandemic (Riva and Wiederhold, 2020; Yang et al. 2021). VR is a digital technology that enables haptic, auditory, and visual multi-dimensional human-computer interaction through a head-mounted display to present the 3D visual environment of a real or imaginary scenes (Thunström et al. 2022). VR technology provides users with simulated experiences, and there are a variety of applications in healthcare, education, commerce, games, entertainment and other industries (Oyelere et al., 2020; Singh et al. 2020).

VR technology has become a hot topic in the research on how to alleviate mental health problems during the COVID-19 pandemic. It has been suggested that mobile device-based VR tools (360-degree video) can be used to combat the psychological pressure caused by COVID-19 (Riva et al. 2020). Virtual forest therapy (VFT) based on VR can aid in coping with the increase of psychological pressure caused by COVID-19 outbreak (Syed Abdullah et al. 2021). An online survey showed that using VR for entertainment could elicit positive emotions during the COVID-19 pandemic (Siani and Marley, 2021). In one study, a VR-based psychological education experience for healthcare professionals was developed to alleviate the psychological pressure of health crisis caused by the COVID-19 pandemic (Pallavicini et al. 2021).

According to a VR industry survey report, the scale of the global virtual reality market is soaring due to the COVID-19 pandemic (it will reach USD 11.64 billion in 2021), and the game industry is likely to lead the VR market in the future (Fortune Business Insights, 2021). Household VR products are gradually becoming a part of daily life. Thousands of users visit VR games every day (Thunström et al. 2022). At the same time, empirical studies have also confirmed that games are the most popular use of virtual VR during the COVID-19 pandemic (Siani and Marley, 2021; Ball et al. 2021). However, there is little research on the effect of household VR products on emotions problems during the COVID-19 pandemic, especially research from the perspective of product users.

With the rapid growth of social media platforms and user-generated content, online user comments have gradually attracted scholars’ attention (Lin et al. 2019; Wang et al. 2021; Yang et al. 2019). Online comments are characterized by a wide range of data sources and a large amount of useful information, and are considered very suitable for research and commercial purposes (Wang et al. 2021). In the area of gaming, online comments usually refer to users’ first-person opinions on relevant game experiences through the game platform, which generally include descriptions of emotions during and after the game (Rogers et al. 2017; Zagal et al. 2009). In order to explore whether household VR products had an impact on the public’s emotions during the COVID-19 pandemic, this study used online comment data of users on the Steam platform to examine comments on VR games and COVID-19. The Steam platform is one of the most popular digital communication platforms for PC games in the gaming industry and one of the largest platforms for VR game distribution (Thunström et al. 2022; Lin et al. 2019). Therefore, research on online comments of users on the Steam platform can obtain representative analysis results in the game domain (Wang et al. 2021). In addition, previous studies using Steam platform data have confirmed that VR games have great potential for treating mental health (Thunström et al. 2022). This study collected data on user comments of VR and COVID-19 on the Steam platform, and analyzed the changes in themes of concern and emotions of relevant user comments over time during the COVID-19 pandemic. The findings of this study provide insights into the impact of household VR products on emotions during the pandemic, offering recommendations for addressing public emotional issues in response to public health emergencies. Additionally, it provides valuable reference for practitioners in the household VR industry and related scholars.

Based on the above, this study attempted to answer the following research questions:

RQ1: During the COVID-19 pandemic, what did users focus on in terms of VR and COVID-19? Did it change over time?

RQ2: What were the semantic-emotion network of user comments?

RQ3: How did the number of comments and types of emotions change over time?

Methodology

Data collection

Prior to data collection, we conducted an analysis of forums related to household virtual reality (VR) products. We found that household VR products are extensively discussed, primarily within the gaming domain. Steam, the world’s largest forum for PC-based game communities, hosts a variety of discussion boards and rich user interactions. Numerous studies have utilized Steam data to explore various topics (Thunström et al. 2022; Lin et al. 2019; Wang et al. 2021).

For this study, we chose to use the Steam Discussion Forums to analyze user emotional responses to VR products during the COVID-19 pandemic. We collected relevant content from discussions in the Steam Community by entering specific search terms into the search box, as outlined in Table 1, and retrieving the corresponding results.

Table 1 Search term.
Full size table

To ensure consistency, only English-language content was collected. The data were gathered on April 14, 2023, with a total of 1,287 user comments collected. The time frame for these comments ranged from January 1, 2020, to December 31, 2022. After manual data cleaning, 36 duplicate comments and two comments containing only URLs (without any comment content) were removed, leaving 1249 valid user comments for analysis.

Emotion and sentiment analysis

Emotion is typically defined as a strong feeling or instinctive reaction, such as joy or disgust (Cambridge University Press, 2025a; Oxford University Press, 2025a). Sentiment, on the other hand, is usually understood as thoughts, views, or feelings about something, such as positive or negative emotions or opinions (Cambridge University Press, 2025b; Oxford University Press, 2025b). In the context of natural language processing (NLP), analyzing emotion-related words like joy or disgust often not only reflects strong feelings but also reveals the positive or negative sentiments of the text’s producer to some extent (Hirschberg and Manning, 2015). In this study, the analysis of emotion helps to uncover the deeper relationships between specific emotions and their semantic network, while sentiment analysis provides insights into the overall trends of positive and negative sentiments. Therefore, we conducted both emotion and sentiment analysis in parallel.

Emotion and sentiment analysis methods are generally categorized into manual analysis and automatic analysis (Boukes et al. 2019; Van Atteveldt et al. 2021). Manual sentiment and emotion analysis is often considered the gold standard, especially when conducted by either expert or non-expert evaluators (Ribeiro et al. 2016). Notably, evaluations performed by both experts and non-experts have been shown to be consistently valid (Snow et al. 2008). However, manual emotion and sentiment analysis is time-consuming and costly (Umer et al. 2018; Zad et al. 2021). Automated analysis addresses these issues of cost and time, but it may sacrifice accuracy compared to manual methods (Boukes et al. 2019; Van Atteveldt et al. 2021). Therefore, manual coding remains more advantageous for smaller-scale tasks. Van Atteveldt et al. (2021) suggested that for optimal accuracy, two coders should be employed for all texts.

For this study, to ensure high accuracy, the first author and the corresponding author manually coded the 1249 comments used in the analysis. Plutchik’s wheel of emotions (Kratzwald et al. 2018; Plutchik, 2001), a commonly used framework in emotion analysis, was adopted for emotion encoding. Eight basic emotions were used in the analysis: anticipation, surprise, anger, fear, trust, disgust, joy, and sadness. Each emotion was coded as a unit, either as a single sentiment token or multiple sentiment tokens. To maintain consistency, only the predominant emotion in each unit was selected for analysis. Among the eight emotions, trust and joy are considered positive, while anger, sadness, fear, and disgust are considered negative. Surprise and anticipation are generally viewed as expressing both positive and negative sentiments (Boon-Itt and Skunkan, 2020). However, in this study, surprise and expectation related to the pandemic and VR tended to be positive expressions. Therefore, in this study, surprise and expectation were treated as positive sentiments. The sentiment analysis was conducted by the two co-authors who independently reviewed each comment. To assess coder consistency, Cohen’s Kappa statistic was used. As shown in Table 2, the Cohen’s Kappa coefficients for both coders were greater than 0.81 for both single and compound sentiments. According to standard interpretations of Cohen’s Kappa, this indicates almost perfect agreement between the coders. These consistency results reinforce the reliability of the coding process and provide strong support for the interpretation of the research findings.

Table 2 Cohen’s Kappa statistic.
Full size table

Semantic network analysis

Semantic network analysis mainly analyzes the relationship between words in a text. In a semantic network, words typically function as nodes, while the relationships between them serve as the links connecting these nodes (Fuhse, 2022; Park and Choi, 2022). This analytical approach is effective for uncovering latent semantic relationships between different topics within textual data, as well as for exploring the connections between emotions and topics in text (Danowski et al. 2021; Park et al. 2019).

We cleaned the collected text data using the SpaCy lemmatization function to perform lexical reduction and eliminate stopwords. We then created a cleaned text data and sentiments matrix and generated a network graph. The nodes in the network represent the cleaned text words and sentiments, while the links indicate the co-occurrence relationships between words and between words and sentiments. The stronger the co-occurrence, the more robust the connection between words and the higher the association between words and sentiments.

In the quantitative analysis of network structure, centrality metrics are commonly used to assess the influence and relationships of nodes within the network. The commonly employed centrality metrics include degree centrality, betweenness centrality, and closeness centrality (Saneinia et al. 2024; Xu et al. 2024). Degree centrality measures the total amount of direct links with other nodes. A higher degree indicates that the node has more direct links, indicating greater direct influence within the network (Park et al. 2022; Freeman, 1978). In this study, the degree centrality of a keyword represents the number of connections it has with other keywords in the network. If two keywords co-occurrence, they are considered connected. Assuming there are n keywords in the network, the degree centrality CD of keyword k is the total number of connections between k and the other n − 1 keywords. The formula for calculation is as follows:

$${C}_{D}\left({p}_{k}\right)=\mathop{\sum }\limits_{i=1}^{n}{x}_{{ik}},i\,\ne\, k$$
(1)

Betweenness centrality measures the extent to which a node plays a “bridge” role in the network. A higher betweenness centrality indicates that the node has a greater role as a connector, and there is a relatively higher probability for other nodes in the network to pass through this node (Freeman, 1978; Gonçalves et al. 2023). In this study, if the number of shortest paths between keywords i and j is denoted as gij, then gij (pk) represents the number of geodesics linking pi and pj that pass through pk. The formula for calculating the betweenness centrality CB of keyword pk is as follows:

$${C}_{B}\left({p}_{k}\right)=\mathop{\sum}\limits_{i < j}\frac{{g}_{{ij}}({p}_{k})}{{gij}},i\,\ne\, j\,\ne\, k$$
(2)

Closeness centrality is a quantitative measure of the length of the shortest path between a node and other nodes in the network. A higher closeness centrality indicates that the node has a relatively shorter distance to other locations in the network (Freeman, 1978; Gonçalves et al. 2023; Lorenzo-Lledó et al. 2023). The formula for calculating the closeness centrality CC of keyword k in this study is as follows:

$${C}_{C}\left({p}_{k}\right)=\frac{1}{[\mathop{\sum }\nolimits_{i=1}^{n}d({p}_{i},{p}_{k})]},i\,\ne\, k$$
(3)

These centrality indices can be used to analyze the influence of nodes (words or sentiments) within the network. In other words, if a word exhibits the highest centrality, it signifies that this word holds greater influence in the context of the period compared to other words. Similarly, if an emotion demonstrates the highest centrality, it indicates that this emotion exerts a stronger influence during that period than other emotions. Additionally, we have employed visualizations for both emotions and words, where the thickness of the lines connecting nodes represents the strength of their co-occurrence frequency. Through these visualizations, we can intuitively discern the potential associations between words and emotions. To gain a deeper understanding of the relationship between topics and emotions, we conducted cluster analysis. Our clustering method uses the Wakita–Tsurumi algorithm for clustering, which determines the structural characteristics of a network by its connection relationships and clusters them (Wakita and Tsurumi, 2007; Kim et al. 2018).

Technical flow diagram

Figure 1 presents our technical flow diagram. We initiated the data collection process on the STEAM platform and subsequently conducted sentiment encoding and text preprocessing on the collected data. Next, we devised our Semantic-Affective Network Model by integrating Emotion and Sentiment Analysis, and Semantic Network Analysis. We constructed the network graph by utilizing text and emotions as nodes and their co-occurrence frequencies as connections. Finally, we analyzed the influence of nodes within the network using centrality metrics, examined the potential relationships between topics and emotions using clustering algorithms, and conducted trend analysis based on the quantity and proportion of each emotion’s associated data.

Fig. 1
figure 1

Technical flow diagram.

Full size image

Results

Content and trends

This study conducted semantic network analysis of user comments on “VR” and “COVID-19” on the Steam platform based on the time of users’ replies on the Steam platform since the outbreak of COVID-19, measuring time in 6-month time units.

As shown in Table 3, from the first half of 2020 to the second half of 2022, the study was divided into six time periods according to the half-year period, and the top 10 keywords with the highest degree centrality in valid user comments in each time period are sorted from high to low.

Table 3 Centrality index of the keywords.
Full size table

In general, “COVID,” “VR,” “pandemic,” “game,” and “like” had the highest centrality among the six phases of user comments. In the first half of 2020, “COVID,” “game,” and “VR” had the highest degree of centrality and betweenness centrality. Next are “pandemic” and “play.” In addition, “people,” “like,” “headset,” “half-life,” and “buy” also show a reasonable degree of centrality, betweenness centrality, and closeness centrality. In the second half of 2020, the highest centrality was found for “COVID” and “VR,” followed by “game” and “pandemic.” In this period, “like,” “go,” “want,” “people,” “time,” and “thing” have high centrality.

For the first half of 2021, “VR” has the highest centrality, followed by “pandemic,” “game,” “like,” “COVID,” “think,” “people,” “go,” “time,” and “get.” In the second half of 2021, “VR” has the highest centrality. “Game”, “COVID,” and “pandemic” have relatively high degrees of centrality and betweenness centrality. This is followed by “work,” “get,” “want,” “play,” “thing,” and “know.” In the first half of 2022, “VR” has the highest centrality. It is followed by “pandemic,” “game,” “COVID,” “play,” “year,” “know,” “thing,” “headset,” and “like.” In the second half of 2022, “VR” has the highest centrality. Next is “COVID,” “game,” and “like,” with a relatively high degree of centrality and betweenness centrality. “Times,” “pandemic,” “get,” “play,” “need,” and “think” had relatively high closeness centrality. Overall, the betweenness and closeness centralities of the words co-occurring at the top of the semantic network performed well.

Theme and the emotion

Figure 2 shows the results of the semantic network thematic clustering of users’ comments in the first half of 2020. The words in cluster 1 were “Halflife,” “Valve,” “buy,” “release,” “new,” “think,” etc. The themes expressed in Cluster 1 mainly reflect the emergence of the new VR game, Half-Life: Alyx, has brought users a completely immersive experience, which is also economically affordable and allows users to release their emotions. This theme is closely related to the user’s “JOY” emotion. The words in Cluster 2 were “headset,” “order,” “need,” “work,” “want,” “spend,” etc. These words reveal the hardships of life during the COVID-19 pandemic, wanting to go to work but being forced to stay at home, and the delays in VR headset orders. This theme is related to the emotion of “ANGER.” The words clustered in Cluster 3 were “sell,” “friend,” week,” “single,” “play,” “make,” “get,” “thing,” “come,” “good,” etc. These words reveal that because of the pandemic, the user could not meet with friends for weeks and could only play the game at home alone. Additionally, the pandemic led to production disruptions, a shortage of VR components, and delayed availability of desired VR headsets. The theme is associated with the emotions of “sadness” and “fear.” In cluster 4, the words “game,” “way,” “have,” “feel,” “job,” “know,” and “issue” express users’ satisfaction with the reliability and quality of virtual reality gaming technology during the pandemic. They also convey a sense of “trust” in the immersive experience and interactivity provided by virtual reality games. Additionally, these words reflect the players’ anticipation for game platform software updates to address issues that arise within the game, as well as their “anticipation” for the ability to work from home for an extended period in the future.

Fig. 2
figure 2

The semantic network clustering of users’ comments in the first half of 2020.

Full size image

The words in cluster 5 were “bad,” “consider,” and “act,” which expressed the users’ sentiment of “anticipation” that the game’s platform will solve the problems in the game and that they will be able to work at home for a long time in the future. The words in cluster 5 further expressed emotions of “disgust” towards the game or the game platform during the outbreak. Examples include delayed delivery of VR headsets (“time,” “run”) and unsatisfactory performance (“bad”). Furthermore, “look,” “fine,” and “price” represent the “surprise” emotion when they discover new features or updates in the VR gaming environment, perceive stable game performance, and encounter lower prices than expected. Other words that emerged from the first half of 2020, such as “wait,” “right,” “money,” “delay,” “day,” “sure,” “lockdown,” and “Steam,” were not clustered with specific emotions but were closely related to various emotional experiences.

Figure 3 illustrates the clustering of comment semantic networks in the second half of 2020. In cluster 1, the words “VR,” “good,” “think,” “halflife,” “gaming,” “problem,” “find,” “feel,” “experience,” etc. express the user’s poor experience with VR games, including technical glitches, poor game optimization, and frustrating game problems. This evokes the emotion of “disgust.”

Fig. 3
figure 3

The semantic network clustering of users’ comments in the second half of 2020.

Full size image

The words “pandemic,” “thing,” “valve,” “wait,” “development,” “reality,” “time,” “bad,” and “future” in cluster 2 suggest that during the pandemic, game development companies (such as Valve) may require more time to ensure games are ready for release. Due to the pandemic’s negative impact on development, more time is required to ensure the game’s quality. Development teams may have faced delays, resulting in delays as users waited for the release of new games or updates on VR gaming platforms. This waiting period can cause feelings of “sadness.” First, words such as “support,” “update,” and “try” express users’ “trust” in the VR game platform. They believe that during the pandemic, the game platform will continue providing stable support services and timely updates, allowing them to enjoy the best VR gaming experience. Second, the words “work,” “headset,” “happen,” and “huge” describe the “joy” users experience from not needing to work during the epidemic and the immense fun the current VR game platform provide. The VR headset experience is delightful and can be shared with other users through the platform. Furthermore, the words “version,” “headset,” “option,” “need,” and “point” reflect the “anticipation” users feel toward VR game updates and development during the epidemic. Users expect further advancements in VR headset technology and more exciting games and features on the platform. They eagerly await more exciting possibilities and opportunities to continue enjoying the VR gaming world after the pandemic ends.

In cluster 5, words such as “new,” “year,” “come,” “sale,” “get,” “look,” “go,” and “friend” express user’s discovery of new features or content in VR games, price discounts or special promotions, and co-op features on VR gaming platforms to play with friends. These elements elicit the emotion of “surprise.”

The words “COVID,” “issue,” “quest,” “user,” “market,” and “current” in cluster 7 describe the serious threats to users’ lives and health caused by the COVID-19 pandemic, technical problems, network lag, or other distress related to the gaming experience on VR platforms. Users experience emotions of “anger” resulting from the surprises associated with the COVID-19 pandemic, the lack of game supply in the VR game market, price increases, or the absence of new content.

In cluster 9, “know,” “likely,” and “video” express users’ uncertainty, contagion, and the consequences of the COVID-19 pandemic that they learn about through communicating with other users and watching videos and information related to the pandemic. This information exchange may trigger emotions of “fear.” Additionally, within VR gaming environments, COVID-19-themed situations, characters, or tasks may trigger “fear” emotions within users.

Other clusters formed in the second half of 2020, such as clusters 3, 6, 8, and 10, showed words that were less directly associated with emotions than those of the other clusters.

The semantic network for the first half of 2021 is shown in Fig. 4. The words “headset,” “right,” “need,” and “different” in cluster 1 reflect the users’ “trust” sentiment regarding VR headset features and performance, proper game strategies for user discussions, and positive responses from the forum community when seeking help or assistance. This evokes “trust” sentiments in cluster 1. In the same cluster, words such as “thing,” “reason,” “expensiveness,” “high,” “console,” “part,” and “run” convey the users’ “anger” emotions about the shortage of VR devices or game accessories due to the pandemic, the high price of VR devices, and VR games that perform poorly or can only be played on specific consoles.

Fig. 4
figure 4

The semantic network clustering of users’ comments in the first half of 2021.

Full size image

In cluster 2, “game,” “demand,” “price,” “available,” and “valve” reflect user’s dissatisfaction with the game content or experience (e.g., expressing dissatisfaction with the lack of content or storytelling in the game), as well as the user’s expectations and needs not being met, the price of the game increasing beyond the user’s means. This elicits “sadness” emotions. Within cluster 2, “people,” “lot,” “sell,” “think,” “afford,” “like,” and other words express users’ concerns about issues in the game (such as malicious behavior, unfair competition, and low-quality content) and their desire to find fun and social connection in games during the COVID-19 pandemic. These problems in the game disrupt their gaming experience and trigger “disgust” in users.

In cluster 3, words such as “COVID,” “get,” “play,” “new,” “room,” and “platform” express how the Covid-19 virus has changed users’ lifestyles, their fear of infection, and the safety of others. The new VR games centered around Covid-19 allow users to feel real fear within the virtual environment, thereby creating “fear” emotions. “Halflife,” “come,” “Steam,” “sure,” “gaming,” and other words express the discovery of new games on the Steam platform, bringing a new VR gaming experience, such as the emergence of the Half-Life games. The exceptional graphics and plot of VR games create excitement among users, generating “surprise” emotions.

In cluster 4, words like “development,” “pandemic,” “hardware,” “ market,” “slow,” “update,” “release,” “computer,” and “user” express the difficulty of development, user hardware limitations, and market uncertainty during the COVID-19 pandemic. Users of VR gaming platforms expressed their “anticipation” emotions in their messages, eagerly awaiting the release of new games.

In cluster 6, “VR,” “great,” “mention,” “nice,” “base,” “good,” “fun,” and “guy” signify that VR gaming platforms have become a haven for many people during the COVID-19 pandemic. Users found immense pleasure and solace in this delightful virtual reality world, offering respite and enjoyment during this difficult time, resulting in a “joyful” mood.

Other clusters formed in the first half of 2021, such as clusters 5, 7, and 8, showed words that were not as distinctly associated with emotions as the words in the other clusters, despite being clustered separately based on their respective themes.

The semantic network for the second half of 2021 is shown in Fig. 5. Words such as “work,” “COVID,” “Facebook,” “ban,” “post,” “sure,” and “day” appeared in cluster 1, reflecting the challenges faced by users during the COVID-19 pandemic, including work-related issues and the frustration caused by daily dilemmas associated with the pandemic. The constant exposure to news and misinformation on social media platforms generated feelings of “disgust” among users. “Work,” “COVID,” “know,” “get,” “issue,” “happen,” and “worker” reflect the user’s constant contact with the risk of COVID-19, fearing that co-workers or themselves become infected, creating tremendous job-related pressure. Additionally, the shortage of vaccines further intensified the “fear” experienced by the users, as they lacked access to sufficient doses to control the outbreak.

Fig. 5
figure 5

The semantic network clustering of users’ comments in the second half of 2021.

Full size image

In cluster 2, “fast,” “home,” “far,” “wait,” “cheap,” “multiplayer,” “headset,” “valve,” “fly,” and “pandemic” depict users’ anticipation of enjoying multiplayer games from the comfort of their homes during the pandemic. They desired inexpensive VR headset devices to engage in multiplayer gaming experiences, providing an escape from reality and the thrill of virtual flying. Users also looked forward to the release of more VR games by Valve.

In cluster 3, words like “pandemic,” “thing,” “base,” “developer,” “development,” “nowadays,” “break,” “small,” “year,” and “ago” convey the numerous challenges and obstacles developers faced during the past two years due to the pandemic and associated restrictions. These difficulties may have resulted in minor delays and issues throughout the development process, generating feelings of “sadness” among users.

In cluster 4, “VR,” “version,” “great,” “hit,” “start,” “experience,” “play,” and other words describe VR gaming platforms that offer version updates and improvements. Older versions of games receive significant upgrades, featuring incredible visual effects and engaging content. Users experience a sense of “surprise” when they begin using these updated versions and are captivated by the new features. Additionally, terms like “market,” “player/user,” “large,” “help,” “upgrade,” and “expect” convey that the VR gaming platform has established a good reputation in the market and that the quality and stability of the gaming platform allows users to upgrade with confidence and expect more satisfying improvements. Users can enjoy these platforms for a long time, thus creating a “trust” sentiment for the platform.

In cluster 5, “well,” “want,” “like,” “especially,” “time,” “support,” “Steam,” “Ubisoft,” and “user” are words that express the immense fun and satisfaction users experienced during the COVID-19 pandemic through Ubisoft’s games and the support of the Steam platform. These games and platforms provided an escape from reality, allowing them to find their own joy in the virtual world and filling their hearts with “joy.”

Only a one-word list appears in clustering 11, expressing the frustration and dissatisfaction of users who cannot complete their task lists due to the limitations of COVID-19, triggering the “anger” emotion in users. Words appearing in clusters 6, 7, 8, 9, and 10 were also clustered separately in the second half of 2021. However, the association between these words and emotions was less pronounced than in the other clusters.

The semantic network for the first half of 2022 is shown in Fig. 6, featuring nine clusters. In cluster 1, the words “face,” “compatible,” “valid,” “issue,” “shortage,” and others express the various challenges faced by users in VR games. These include uncomfortable scenarios, headset equipment problems, technical problems or failures with VR headset equipment or game platforms, and shortages of VR equipment or timely game content updates. Users experience “disgust” due to these issues. Words like “hope,” “chip,” “end,” and “reality” express users’ fear experienced during the COVID-19 pandemic. They were worried that electronic chips or hardware problems influenced by the epidemic might lead to VR games’ malfunction or poor performance. Additionally, the realistic sense of reality in games blurs the line between the virtual and real world, resulting in an emotion of “fear.”

Fig. 6
figure 6

The semantic network clustering of users’ comments in the first half of 2022.

Full size image

In cluster 2, words such as “VR,” “Halflife,” and “know” express the users’ immersive experience with Halflife games. Users are amazed by the immersive nature of the Halflife game, describing it as an incredible new world and reflecting their emotion of “surprise.” Words like “Halflife,” “story,” “afford,” “give,” “miss,” and “unresolved” reveal users’ “anger” about the unresolved Halflife storyline. Users also experienced frustration when their characters could not jump to key locations because of VR equipment failure, intensifying their “anger” as they had invested much time and effort into overcoming this challenge. Words like “link,” “down,” “height,” “fantastic,” and “want” describe users’ sense of “anticipation” toward VR games and the future development of the game platform. Users expect the VR game platform to achieve their desired vision and believe in its potential to deliver even more exciting and immersive experiences, taking VR gaming to the next level.

Cluster 3 includes words like “year,” “PC,” “moment,” “thing,” “good,” “release,” “version,” “make,” “friend,” and “Steam, depicting users’ experiences on the VR gaming platform on PC during the COVID-19 pandemic. Users showcase their friendship and love by reminiscing, adapting plans, experiencing moments of grief, and exploring versions and stories that evoke feelings of “sadness.”

In cluster 4, words like “COVID,” “play,” “people,” “home,” “flight,” “minute,” and “world” reflect the escapism and joy users find in the virtual environment of VR games, which is completely different from the real world. Users can temporarily forget their worries, interact with others from the comfort of their homes, explore virtual worlds together, and immerse themselves in the “joy” of the game.

Cluster 5 has “headset,” “pandemic,” “start,” “real,” “get,” “cool,” and “spend,” describing how VR gaming platforms provided real experiences, timely issue resolution, user interaction, user-friendly interfaces, and high-quality content and VR headset devices during the COVID-19 pandemic. This instilled “trust” among users, convincing them that the VR platform was reliable and worth their time and money.

Other clusters in the first half of 2022, such as clusters 6, 7, 8, and 9, contained words clustered separately. However, the association of these words with emotions was less pronounced than in the other clusters.

The semantic network for the second half of 2022 is shown in Fig. 7. In cluster 1, words like “delay,” “cost,” “lack,” “support,” “time,” “like,” “development,” “market,” “gaming,” and “high” reflect the “disgust” users experienced during the COVID-19 pandemic. This was caused by game delays, increased costs, limited game variety, inadequate support, technical failures, and poor-quality development. These factors contributed to feelings of “disgust.”

Fig. 7
figure 7

The semantic network clustering of users’ comments in the second half of 2022.

Full size image

In cluster 2, words like “game,” “want,” “team,” “play,” “pandemic,” “story,” “main,” “go,” “future,” and “Half-life” reveal that in the COVID-19 pandemic, users find enjoyment and build trust with other users through teamwork and practical skills on VR gaming platforms. The fun and trust are associated with the Half-life series characters and stories, bringing users excitement and satisfaction. Despite economic challenges, users find joy and happiness in escaping reality through the game without incurring significant costs. Users’ emotions of “trust” and “joy” are reflected in this cluster.

In cluster 3, words like “VR,” “people,” “develop,” “pay,” “use,” “feel,” “real,” “open,” and “chat” express that, during the COVID-19 pandemic, users could engage with others through VR gaming platforms. However, the inability to authentically feel real-world human interactions and emotional connections may have evoked “sadness” for users in this virtual environment.

In cluster 4, words like “developer,” “MMO,” “virtual,” “start,” “Valve,” “title,” “look,” and “gaming” describe the VR MMO game’s ability to bring “surprise” amid the present reality. Users can learn new skills and spend quality time through VR games.

In cluster 5, words like “good,” “GPU,” “luxury,” “need,” “headset,” “mean,” “item,” and “big” express the emotions of “anticipation” among gamers during the COVID-19 pandemic. Users anticipate high-quality games, powerful graphics performance, and a unique experience meeting their personal needs. They desire a more realistic feel facilitated by high-quality headsets and the enjoyment of massively multiplayer games or significant updates.

In cluster 7, words like “get,” “release,” “say,” “new,” “community,” “consider,” and “dev” show the frustration caused by delays due to the COVID-19 pandemic and the lack of access to new games. Users may feel “anger” when encountering problems in the game or wish to make suggestions but cannot communicate directly with developers or receive timely responses.

For clusters 6 and 8, the words that appeared were clustered separately, but the association of these words with emotions was not as clear as in the other clusters.

Changes in comment volume and emotion

The user comments and distribution of emotions from the first half of 2020 to the second half of 2022 (in half-year increments) are shown in Fig. 8. In the first half of 2020, the user comments were 454, with “trust” being the most frequent emotion, with 129 comments. The second was “anticipation,” with 127 comments. The third most common was “sadness,” with 89 comments. These are followed by “disgust,” “anger,” “joy,” “surprise,” and “fear,” with 84, 61, 51, 127, and 24 comments, respectively. In the second half of 2020, the user comments significantly decreased to 274. Unlike the first half of 2020, “anticipation” became the most frequent emotion, with 101 comments. The second most frequent emotion was “trust,” with 73 comments, and the third was “disgust,” with 64 comments. These were followed by “sadness,” “joy,” “anger,” “fear,” and “surprise,” with 58, 43, 40, 21, and 18 comments, respectively. The user comments decreased in the first half of 2021, with 209 comments. Comments with the “anticipation” emotion were the most frequent, but the number was reduced to 87. Next were “trust,” “sadness,” “joy,” and “surprise,” with 59, 44, 41, and 37 comments, respectively. These were followed by “anger” (37), “disgust” (27), and “fear” (6). In the second half of 2021, the user comments decreased to 138. Comments with the “anticipation” emotion were still the most frequent, but the number was reduced to 76. The second was “ trust,” with 47 comments, and the third was “sadness,” with 30 comments. These were followed by “surprise,” “joy,” “disgust,” “anger,” and “fear,” with 20, 18, 17, 12, and 5 comments, respectively. In the first half of 2022 and the second half of 2022, the user comments will continue to decrease to 103 and 70, respectively. In the first half of 2022, the most frequent emotion in comments was “sadness,” with 32 comments. The next most frequent was “anticipation,” with 27 comments. The third and fourth were “anger” and “disgust,” with 18 and 17 comments. These were followed by “joy,” “trust,” “fear,” and “surprise,” with 13, 11, 9, and 8, respectively. In the second half of 2022, the comments expressing “anticipation” emotion once again became the most frequent emotion, which is 33, while “ trust “ and “ sadness “ appeared 15 and 14 times. Next were “disgust,” “surprise,” “anger,” “joy,” and “fear,” with 10, 5, 4, 4, and 0 comments, respectively.

Fig. 8
figure 8

The number trends of the distribution of emotions.

Full size image

As shown in Fig. 9, the proportion of positive emotions was higher during the six periods than that of negative emotions. In each period, only the proportion of negative emotions in the fifth period was higher than that of positive emotions. Among the eight emotions, the proportion of “anticipation” emotion was the highest, reaching the highest in the second half of 2022 at nearly 40%. However, the proportion of “anticipation” was lower than “trust” in the first half of 2021 and in the first half of 2022, lower than that of “sadness.” The proportion of “trust” and “sadness” in the six periods fluctuated, with the most significant in the first half of 2022. In that period, the proportion of “trust” decreased by half compared to other periods in the first half of 2022, while the proportion of “sadness” increased by half compared to other periods. The proportion of “anger” remained around 10% in the first three periods but suddenly decreased to 5% in the second half of 2021. Subsequently, it surged to 13% in the first half of 2022 and again reduced to 5% in the second half of 2022. The proportions of “disgust” and “fear” hardly fluctuated in the first two periods. In the third and fourth periods, they decreased to half of the proportions in the first two periods. After a surge in the fifth period, they decreased again in the sixth period, especially the proportion of “fear,” which reached zero. The other two emotions exhibited a continuous fluctuating trend over the six periods; however, the range of fluctuations was not extensive.

Fig. 9
figure 9

The percentage trends of the distribution of emotions.

Full size image

Discussion

This study investigates the dynamic changes in user concerns during the COVID-19 pandemic by analyzing user comments related to household virtual reality devices on the Steam platform. A combined approach of sentiment analysis and topic analysis was adopted to construct a topic-sentiment network model, providing an in-depth examination of the relationship between sentiment types and topics in user comments, as well as the temporal trends of sentiment changes. Textual discussions in online forums are not only a form of social interaction but also influence others’ emotional responses through the expression of emotions (Reis and Collins, 2004). Specifically, emotional expression and the social context are intertwined, as individuals’ emotional cues can elicit inferences and reactions from others (Van Kleef, 2010). Against this backdrop, social interaction and emotional responses offer valuable perspectives for understanding changes in user emotions during the pandemic.

Trends in emotion and themes during the pandemic

This study reveals the trends in emotional and thematic changes in user comments during the pandemic. Specifically, during the COVID-19 pandemic, the top four words with the highest network centrality in user comments, excluding “COVID-19,” “VR,” and “pandemic,” are all related to “game” and “like.” Despite the ongoing context of the pandemic, the immersive experience of VR games became the focal point of user discussions, to the extent that it somewhat shifted attention away from the pandemic itself. Overall, discussions about VR gaming experiences were more frequent than those about COVID-19, which is consistent with previous research on the application of VR in therapeutic settings (Smits et al. 2020). This suggests that the immersive nature of VR games provided users with a psychological refuge, helping them temporarily escape the anxiety and stress caused by the pandemic (Smits et al. 2020). Particularly in the early stages of the pandemic, users’ enthusiasm for VR reflected their need for emotional regulation in the face of uncertainty (Sanderson et al. 2020).

Anticipation and trust: dominant user emotions over time

Through semantic network analysis, this study further reveals the emotional trends in user comments across six time phases. Throughout all phases, “anticipation” and “trust” were the most prominent positive emotions, with the proportion of “anticipation” remaining particularly high. This suggests that despite the uncertainty and lifestyle changes brought by the pandemic, users maintained a hopeful outlook for the future, particularly with regard to the development of VR technology and applications. In the early stages of the pandemic, users’ “anticipation” focused on the new experiences and free time they hoped to gain from VR under the home-office model. However, as the pandemic control measures were extended, this emotion gradually shifted to anticipation for the post-pandemic recovery period. By the first half of 2021, users generally recognized that COVID-19 had accelerated the development of the VR market, further fueling their expectations for improvements in VR games and the future of the VR industry. Moreover, the emotion of “trust” was prevalent across all stages. In the early stages of the pandemic, users mainly expressed trust in VR platforms, technology, games, and devices. In subsequent phases, as users adapted to VR, their trust was combined with positive experiences. For instance, using Steam-VR to watch movies with friends helped alleviate feelings of loneliness and depression caused by the inability to meet in person, indicating that VR was not just a tool for entertainment but also a medium for social and emotional connection. This finding is consistent with existing research, which shows that VR, as a social tool, helped users maintain emotional bonds with others during the pandemic (Riva et al. 2020; Marston and Kowert, 2020). Emotional expression conveys not only personal feelings but also emotional intentions (Van Kleef, 2010). This was reflected in the semantic-emotion network model of this study. The research indicates that VR usage can convey positive emotions such as “joy,” which is consistent with findings that VR helped alleviate stress for frontline healthcare workers (Labrague, 2021; Sun et al. 2021), and also aligns with studies on reducing psychological pressure among isolated individuals and university students (Xu et al. 2020; Chen et al. 2020).

Shift in negative sentiments and long-term adaptation

While positive sentiments dominated overall, the expression of negative sentiments also exhibited notable temporal variation. In the first half of 2020, users’ negative sentiments were primarily centered around “sadness,” “anxiety,” and “stress” caused by the pandemic, reflecting the widespread social unease and psychological distress at the beginning of the pandemic. However, by the second half of 2020, the focus of negative sentiments gradually shifted to technical issues with VR games. Specifically, user frustrations arose from VR game malfunctions, lagging operations, and delays in the production of VR devices due to the pandemic. This shift suggests that as the pandemic persisted, users’ concerns moved from the social uncertainties of the pandemic itself to specific problems related to VR technology. This finding aligns with research on the use of VR in stress and anxiety relief (Riva et al. 2020), and suggests that as the pandemic progressed, users’ expectations for VR increasingly focused on technological improvements rather than solely relying on its emotional regulation capabilities (Riva et al. 2020). Finally, regarding changes in the volume of comments, this study found that since the outbreak of the pandemic, the number of user comments gradually decreased, particularly in the first half of 2022, when the proportion of “anticipation” in comments significantly declined. This phenomenon may be related to the users’ increasing adaptation to the pandemic. As the pandemic entered its prolonged phase, users’ emotional responses stabilized, and both positive and negative sentiments became more neutral, with fewer instances of intense emotional fluctuations. This trend is similar to emotional changes observed in the context of online education and remote work, suggesting that after long-term adjustments to pandemic-related lifestyle changes, users gradually reached an emotional adaptation (Chakraborty et al. 2021; Mishra et al. 2020).

Theoretical contribution

This study combines semantic network analysis with emotion analysis to explore the emotions associated with household VR products, innovatively constructing a semantic-emotion network model. This model not only enhances the granularity and depth of emotion analysis but also provides methodological support for analyzing the interactive mechanisms between VR technology and user emotions. Furthermore, this study reveals the complex relationship between emotional changes and VR usage scenarios, thereby extending the theoretical framework of virtual reality in the context of social interaction and emotional regulation.

Practical contribution

The findings of this study may offer valuable practical insights for both industry and policymakers. From a policy perspective, governments and public health organizations should recognize the emotional support functions of virtual reality technology in times of social crisis. By supporting the development and dissemination of relevant technologies, governments can provide more effective emotional support interventions for the public, particularly in special contexts such as quarantine and remote work. In addition, policymakers should encourage the cross-disciplinary application of VR technology in fields like education, entertainment, and healthcare, promoting its widespread use in public life. From a business perspective, by revealing the emotional fluctuations experienced by users during their interactions with household VR products, companies can gain a better understanding of user needs, optimize product design, and enhance user experience. Moreover, this study provides valuable insights for industries such as gaming, mental health, and education, particularly in crisis situations, further validating the potential of VR products as tools for emotional regulation.

Limitations and future research directions

This study primarily relies on user comments from the Steam platform, which may introduce some platform bias. Additionally, due to the lack of demographic information provided by the platform, individual differences and their potential impact on the results were not considered. These are limitations of the study. This research focuses primarily on emotional changes related to VR gaming; future studies could further explore the emotional impact of other types of VR applications (e.g., healthcare, education) on users, particularly examining differences across various domains before and after the pandemic. Furthermore, given the profound impact of the pandemic on societal mental health, future research could also consider long-term tracking of user emotions to investigate the role of VR products in psychological recovery and social rebuilding post-pandemic.

Conclusions

Based on VR-related comment data, this study analyzed the trend of emotion changes during the COVID-19 pandemic and uncovered the relationship between VR-related content and emotion during this period. The findings reveal that, during the pandemic, anticipation and trust dominated the emotions of VR users, with the use of virtual reality serving as one of the sources of positive sentiments. Additionally, a shift in negative sentiments was observed, reflecting a phenomenon of long-term adaptation. Users’ negative sentiments transitioned from content related to the pandemic to technical issues associated with VR use, and over time, emotional adaptation occurred. This study innovatively constructed a semantic-emotion network model to analyze the interactive mechanisms between VR technology and user emotions, enhancing the granularity and depth of emotion analysis while revealing the complex relationship between emotional changes and VR usage scenarios. Furthermore, the study suggests that public health departments recognize the emotional support functions of VR technology during social crises and implement appropriate virtual reality-based emotional support interventions for situations such as quarantine and remote work.