Social media sentiment: How people online are feeling, and responding to, the COVID-19 emotional rollercoaster

Greater Melbourne and Mitchell Shire have returned to stage three restrictions, and we find ourselves, once again, going online to connect with extended family, friends and our communities.

The increased reliance on online spaces to connect, express ourselves, and seek out information and news about the pandemic during stay-at-home restrictions, or lockdowns in some places, makes the online realm an interesting site to gauge the sentiment of our community during this crisis.

Over the course of the pandemic, I've been conducting topic modelling and analysing the sentiment of social media posts, comments and tweets across several social media platforms to develop an understanding of the needs and concerns users are having at various stages of the crisis. I've specifically been looking at certain hashtags, forums and groups that align with various regional and metropolitan areas in Victoria, Queensland, NSW, WA and Tasmania as part of a larger project.

At the beginning of the pandemic, conversations were dominated by humorous posts about toilet paper, concerns about the cancellation of flights, and speculation about what the novel coronavirus was, how it's transmitted, and its incubation period.

As restrictions were implemented, these conversations shifted. Users began discussing Centrelink, superannuation, parking, restrictions, hospitals, police, and schools, among other topics.

The sentiment around hospitals became increasingly negative as the pandemic escalated, with complaints about limited access to GPs in regional areas, as well as the suspension of elective surgery.

Sentiment regarding the police was very negative, which isn't atypical, as crime reports trigger negative results due to words such as "murder". However, a closer look at the conversations demonstrates much of the negative sentiment was related to complaints about the police response to enforcing restrictions, as well as a backlash to police speed cameras in WA.

In Melbourne, there was positive sentiment about the police, which emerged in response to the deaths of four Victorian police officers in a traffic accident in April. The positive sentiment represented was about raising funds for the four officers’ families, and honouring their legacies.

Centrelink and JobKeeper became frequent topics of conversations. Interestingly, on average these conversations were positive in sentiment.

Making sense of sentiment analysis

It's understood by linguists that human expressions can contain multiple or mixed sentiments. To capture mixed sentiment, each text (comment, tweet, post etc) is given a positive, neutral and negative score. VaderSentiment also provides a "compound" score that calculates an overall sentiment score for the text.

Crucially, the sentiment within a comment may not be in relation to the topic of that comment. Although there were many complaints about the long lines or waits at Centrelink, users maintained a positive or joking tone in these posts.

For example, one user posting a question about youth allowance payments continued by saying: “I would prefer not to ring them and spend three hours on the phone if possible haha. Also, gorgeous day today.”

This resulted in a strong positive sentiment.

Similarly, users would frequently go online to seek help with the application process for JobKeeper or information about how their current payments would be affected. Many users came forward to provide assistance, advice and encouragement, resulting in many comments with positive upturns such as “happy to help”, “thanks so much!”, “appreciate it mate”.

These findings of an overall positive sentiment are indicative of a community coming together in a time of need to help each other out.

A similar result emerges from conversations happening in Melbourne and Victoria as we went back to stage three restrictions. The illustration below shows some of the key topics of conversation, and the associated sentiment – the bubbles are sized based on frequency, while the colour reflects the strength of the positive or negative sentiment.

Commonly, discussions that began as other topics inevitably moved back to focus on the virus, with conversations about masks dominating.

Read more COVID-19: Face masks are no silver bullet, but they're important

Many of the discussions involved users informing one another about the correct way to wear masks, what masks are best, and recommending places and companies to purchase quality masks.

Users also commonly shared photos of their local area, views of the city and sunsets, and photos of their pets as they spent more time at home, contributing to a positive sentiment online.

Read more: View from my window: Sharing lockdown in pictures

Negative sentiments emerged when users expressed concern about telephone and cyber scams, experiences with racism (especially against Chinese or Muslims), and complaints about people breaching restrictions.

There was also concern expressed about the experiences of those in public housing towers, sharing stories from residents, and complaints about how the lockdown was managed.

Read more: Nine Melbourne tower blocks put into 'hard lockdown' – what does it mean, and will it work?

Finally, there were many discussions about the security guards breaching protocols, but these conversations often included a humorous tone with many jokes, boosting the sentiment to a more positive score.

Within this dataset, some users posted cries for help online, revealing they were struggling, and unsure they would make it through the “second wave” as a result of losing their job at the start of the pandemic.

These expressions of acute anxiety and depression were met with outpourings of sympathy and offers of help, with many users responding with care, concern and encouragement, sharing links to mental health resources, offering suggestions for potential work opportunities, information about how to access superannuation early, as well as general hope for the future, and solidarity with frequent statements such as “we are all in this together”.

The overall positive nature of the discussions online is a reflection not of the happiness of users, but of how people are coming together to provide support during a time of crisis. It shows that genuine and productive civic discourse can, and is, happening online in a time of need.