You are viewing this site in staging mode. Click in this bar to return to normal site.

Using Twitter to Track Public Opinion in the UK About Mental Health During the COVID-19 Pandemic

Christopher Marshall*, Kate Lanyi*, Rhiannon Green, Georgina C Wilkins, Savitri Pandey, Dawn Craig 

 

NIHR Innovation Observatory, Newcastle University 

 

*Corresponding author: chris.marshall@io.nihr.ac.uk

 

Abstract 

We examined the use of Twitter as a potential means to monitor, extract and analyse useful insight from the UK public in relation to the concern and impact of the COVID-19 pandemic on mental health. A search strategy comprising a list of terms relating to COVID-19, the lockdown and mental health was developed and used to search for relevant tweets. We used specialist text analytics and artificial intelligence platform to analyse the tweets in terms of volume, sentiment, and key topics of discussion. Two independent researchers carried out qualitative document analysis to further explore and expand on the results generated by the platform. We identified and collected 380,728 tweets from 184,289 users in the UK over a nine-week period (30 April to 4 July 2020). The highest volume occurred in the second week with 61,874 tweets. The lowest volume occurred in week eight with 21,578 tweets. The platform calculated a total sentiment score of 52, suggesting a slightly positive overall sentiment across all of the tweets. Various topics of discussion with both positive and negative underlying sentiments were identified and summarised. The results suggest that this is a potentially useful and efficient way of pulling out insights from the ‘public voice’ relating to public health matters. However, there are still various limitations to consider concerning the technology and representativeness of the data. Future work to validate and further explore the value of this type of research is recommended. 

1 Introduction 

COVID-19 was identified as a new type of coronavirus in early January 2020.1 Since then, the disease has rapidly spread to many parts of the world, including the UK where the first outbreak was reported on 31 January 2020. COVID-19 was declared a global pandemic on 11 March.2,3 

The COVID-19 pandemic is having a profound effect on mental health. In a key position paper published in June 2020, the authors explored the psychological, social, and neuroscientific effects of COVID-19 and set out a series of immediate priorities and longer-term strategies for mental health research 4. In the paper, one of the immediate research priorities set out is ‘surveillance’. In particular, the authors suggest that we must find useful ways to monitor and analyse data on the mental health effects of the COVID-19 pandemic across the whole population, as well as vulnerable subgroups 4. 

With over 300 million monthly users, Twitter is one of the most popular social media platforms currently available. Twitter is a free micro-blogging service that enables its users to send and read each other’s “tweets” (i.e. short messages limited to 280 characters). Over time, Twitter has been increasingly used as a tool and data source for health-related research, with the potential for offering a more efficient means of data collection over traditional survey methods 5. In particular, Twitter has been used to monitor, track trends, and disseminate health information during past viral pandemics 6-9. Further, previous studies have leveraged Twitter data for the assessment of public sentiments, attitudes and opinions concerning health-related issues 10,11.  

In this work, we examine the use of Twitter as a means to monitor, extract and analyse useful insight from the UK public in relation to the concern and impact of COVID-19 on mental health. 

 

2 Methods 

2.1 Data collection 

We developed a search strategy comprising a list of terms related to COVID-19, the lockdown and mental health to search (or scrape) for relevant tweets (see Figure 1). 

Figure1.Searchstrategyforrelevanttweets.docx 

The search terms were identified through discussion within the research team and scanning the background literature. Once the strategy had been agreed, we began prospectively searching for and scraping relevant tweets using Twitter’s advanced search application programming interface.  

In this article, we are reporting the findings from our analysis of relevant tweets scraped over a nine-week period, from 30 April 2020 (the date we operationalised our prospective search for tweets) to 04 July 2020 (the date the UK lockdown was considerably eased). 

 2.2 Data analysis 

We used a specialist text analytics platform to initially analyse the tweets, followed by qualitative document analysis to further explore and expand on the results. 

The platform is described by its developers as a text analysis and insight platform utilising artificial intelligence techniques1. In particular, the platform supports topic and sentiment analysis. Qualitative document analysis is a methodological approach enabling the evaluation of printed and electronic materials and has been useful in previous studies for examining social media data 12,13. 

Using the platform, we were able to track and determine the tweet frequency across the UK, identify key emerging topics of discussion, and explore the overall trend in sentiment (i.e. positive, neutral, or negative). All collated tweets were anonymised. 

Two researchers (KL and RG) ran the initial analysis using the platform and carried out qualitative document analysis independently. The findings were discussed by all of the authors and consolidated to form a final set. 

 

3 Results 

3.1 Volume of tweets 

Using our search strategy, we captured 380,728 tweets from 184,289 users in the UK from 30 April 2020 to 4 July 2020. Figure 2 presents the flow in volume of tweets over the study period. 

Figure2-Volumeoftweetsfrom30Aprilto4July2020.docx

The highest volume occurred week commencing (w/c) 4th May with 61,874 tweets. The lowest volume was observed w/c 22 June 2020 with 21,578 tweets. Discounting w/c 27 April (as this week did not contain a full weeks’ worth of tweets), the data shows an overall downward trend in the volume for the first eight weeks. However, in the ninth week (w/c 29 June), a sharp increase is observed. 

 3.1.1 Filtering by mental health issues 

 Keyword filters were set up around the topics ‘anxiety, ‘depression’, ‘loneliness’ and ‘stress’. Table 1 summarises the volume of tweets identified using each filter, which was run on the overall dataset.  

Table1.Keywordfiltersformentalhealthissues.docx 

Figure 3 presents the flow in volume of tweets around each keyword filter over the study period. 

Figure3flowinvolumeoftweets.docx

For tweets identified using the ‘anxiety’ filter, the highest volume was observed w/c 27 April with 4,090 tweets. The lowest volume was observed w/c 8 June with 1,702 tweets. Overall, the data shows a steady downward trend in the volume of tweets over the study period. 

For tweets identified using the ‘depression’ filter, the highest volume of tweets occurred w/c 4 May with 5,085 tweets. The lowest volume of tweets was observed w/c 22 June with 899 tweets. Overall, the data shows a downward trend in the volume of tweets. However, spikes in volume occurred w/c 18 May and 15 June. 

For tweets identified using the ‘loneliness’ filter, the highest volume of tweets was observed w/c 15 June with 1,890 tweets. The lowest volume of tweets occurred w/c 29 June with 770 tweets. Throughout the study period, the number of tweets was falling over an initial three-week period (w/c 18 May to 1 June) before sharply rising again over the following three-week period (w/c 1 June to 15 June), peaking w/c 15 June. In the final three weeks, the volume was falling again. 

For tweets identified using the ‘stress’ filter, the highest volume of tweets occurred w/c 4 May with 2,545 tweets. The lowest volume of tweets was observed w/c 29 June with 1,077 tweets. Overall, the data shows a steady downward trend in the volume of tweets over the study period. 

3.2 Sentiment analysis 

A total of 380,728 tweets were identified and analysed for sentiment using the platform. Of these: 

  • 70,262 (18%) tweets were identified as having positive sentiment 
  • 272,906 (72%) tweets were identified as having neutral sentiment 
  • 37,560 (10%) tweets were identified as having negative sentiment 

The platform calculated a total sentiment score of ‘52’. A score less than 50 would suggest negative sentiment, whilst a score higher than 50 would suggest positive sentiment. Therefore, this score (52) suggests a slightly positive overall sentiment across all of the tweets. 

Figure 4 visualises the weekly flow in sentiment over the study period. 

 Figure4Volumeoftweetsfrom30Aprilto4July2020.docx

 Overall sentiment remained positive throughout the study period, with the total score staying consistently above 50. Highest sentiment occurred w/c 15 June (53.6). Lowest sentiment occurred w/c 25 May (51). 

Over the first four weeks, sentiment was quite varied. In the three-week period from w/c 25 May to 15 June, there was a consistent and steady rise in sentiment before a sharp decrease in the final two weeks. 

 

3.3 Topic analysis 

 Before analysing and exploring the key topics of discussion within the data, it was necessary to apply a final keyword filter. This filter comprised a series of keywords associated with mental health issues, which would help ensure a more relevant and cleaner dataset for the analysis (see Figure 5). 

 Figure5.Mentalhealthkeywordfilterfortopicanalysis.docx

Applying this filter reduced the total number of tweets from 380,728 to 187,566. Therefore, the results of this topic analysis are based on the 187,566 collected and filtered tweets. 

In this section, the results are broken down into, roughly, three-weekly chunks of data, as follows: 

  • Section 3.3.1 presents the results for tweets posted w/c 30 April to 24 May 
  • Section 3.3.2 presents the results for tweets posted w/c 25 May to 14 June 
  • Section 3.3.3 presents the results for tweets posted 15 June to 4 July 

 

3.3.1 Results of topic analysis for weeks 1 to 3 

This section presents the results of a topic analysis consisting of a sample of 93,489 tweets posted from 30 April to 24 May. Table 2 summarises a selection of notable events that happened during this time period. 

 Table2.Notableeventsthatoccurredfrom30Aprilto24May.docx

 

Table 3 presents the top 10 most discussed topics during this time period. 

Table3.Top10mostdiscussedpositiveandnegativetopics.docx

3.3.1.1 Summary of key topics discussed with positive sentiment 

‘Mental health’ emerged as a key topic of discussion underpinned with positive sentiment. Users recognised the ‘coronavirus pandemic’ was going to be a very ‘difficult time’, particularly in relation to people’s mental health. A selection of people took to twitter to highlight the positive impacts that the lockdown had been having on their mental health. In particular, some people were discussing how they had surprised themselves at their ability to adapt to the lockdown by creating good habits and routines to stay healthy. 

There was positive discussion around ‘mental health awareness week’ (18 to 24 May), where the theme this year was ‘kindness’. During this week, people were sharing on twitter helpful ways (e.g. videos, charities, helplines, exercise regimes, healthy eating advice) others could look after their mental health, particularly those in ‘self-isolation’. 

There was considerable discussion around ‘vulnerable people’. There was positive discussion from people around ensuring those who may be particularly vulnerable in relation to mental health will get the support they need within their communities. Specific groups highlighted included the elderly, those self-isolating, those from LGBT groups and those living alone. In particular, many people expressed how important it was to one’s mental health to stay connected with ‘family and friends’ during this period. 

People were also tweeting their gratitude for the work that individuals and organisations were doing to support mental health. 

A selection of example tweets classified with positive sentiment during this time period is shown in Figure 6. 

Figure6.Exampletweetsfromweeks1to3withpositivesentiment.docx

3.3.1.2 Summary of key topics discussed with negative sentiment 

‘Mental health’ also emerged as a key topic of discussion underpinned with negative sentiment. People were tweeting and discussing how much their mental health had deteriorated during the lockdown. People were describing specific issues they had been facing around financial stress, employment worries and loneliness. Some people described how they had increased their alcohol intake to try and cope. Others mentioned how they had stopped keeping up with the news as it was causing too much anxiety and distress. 

Following an address by the UK Prime Minister on 10 May, there was considerable discussion around ‘lifting the lockdown’. Opinion on Twitter around lifting the lockdown was divided. Some expressed how the lockdown should be lifted due to the damage it is causing to mental health. However, others were worried that lifting the lockdown too early would worsen their anxiety. 

A selection of example tweets classified with negative sentiment for this time period is shown in Figure 7. 

Figure7.Exampletweetsfromweeks1to3withnegativesentiment.docx

 

 3.3.2 Results of topic analysis for weeks 4 to 6 

 This section presents the results of a topic analysis consisting of a sample of 43,010 tweets posted from 25 May to 14 June. Table 4 summarises a selection of notable events that happened during this time period. 

 Table4.Notableeventsthatoccurredfrom25Mayto14June.docx

Table 5 presents the top 10 most discussed topics during this time period. 

 Table5.Top10mostdiscussedpositiveandnegativetopics.docx

3.3.2.1 Summary of key topics discussed with positive sentiment 

‘Mental health’ was still a key topic of discussion underpinned with positive sentiment. People were commenting that the pandemic (and resulting lockdown) was raising even more awareness in relation to mental health and its importance. Some people described how they had been using ‘lockdown life’ as a positive opportunity to work on their mental health. People also continued to share useful resources on mental health, particularly in relation to ‘vulnerable people’. 

Following the UK Government’s easing of lockdown in England on 01 June 2020, ‘life after lockdown’ became a topic of considerable discussion. Some people described how the lockdown easing was having a positive impact on their mental health, particularly in mitigating loneliness. 

A selection of example tweets classified with positive sentiment during this time period is shown in Figure 8. 

Figure8.Exampletweetsfromweeks4to6withpositivesentiment.docx

 3.3.2.2 Summary of key topics discussed with negative sentiment 

‘Mental health’ also continued to be a key topic of discussion underpinned with negative sentiment. People discussed how prolonged feelings of isolation due to lockdown had considerably impacted their mental health.  

Sleep was also a common topic of discussion. Many people were describing how they had been struggling to sleep at night due to stress and anxiety. Some people also described that they had been experiencing sleep walking and unpleasant dreams. 

Some people expressed that the UK Government’s steps to ease lockdown were causing more stress and anxiety than being under lockdown conditions. In particular, people voiced concerns that the lockdown was ending too soon and without adequate protection in place for vulnerable groups. Further, some people were anxious about leaving the house and using public transport again. 

A selection of example tweets classified with negative sentiment during this time period is shown in Figure 9. 

Figure9.Exampletweetsfromweeks4to6withnegativesentiment.docx

 

 3.3.3 Results of topic analysis for weeks 7 to 9 

This section presents the results of a topic analysis consisting of a sample of 51,362 tweets posted from 15 June to 4 July. Table 6 summarises a selection of notable events that happened during this time period. 

Table6.Notableeventsthatoccurredfrom15Juneto4July.docx 

Table 7 presents the top 10 most discussed topics during this time period. 

 Table7.Top10mostdiscussedpositiveandnegativetopics.docx

 3.3.3.1 Summary of key topics discussed with positive sentiment 

During this time period, a tweet suggesting suicide had risen by “200% since lockdown” was shared widely by users. The tweet also contained contact details for registered UK charity, Samaritans, urging people to reach out for support if needed. This viral tweet resulted in the platform recognising ‘suicide lockdown’ as the number one topic of discussion. It was later revealed that the content of this tweet was inaccurate. 

Key topics of discussion from the previous six weeks were still prevalent during this period. People continued to share positive stories in relation to improving/maintaining their mental health, useful resources, support for ‘vulnerable people’, and gratitude for key workers/organisations. ‘Family and friends’ also resurfaced as a popular topic of discussion. 

Some people were also discussing ‘working from home’ and sharing tips and strategies around best practice. 

A selection of example tweets classified with positive sentiment during this time period is shown in Figure 10. 

Figure10.Exampletweetsfromweeks7to9withpositivesentiment.docx

3.3.3.2 Summary of key topics discussed with negative sentiment 

A community-driven ‘copy and paste’ tweet campaign (‘friends just copy’), concerning how difficult the lockdown had been for those with depression, was identified by the platform as a key negative topic of discussion. 

As with previous weeks, people continued to share how badly their mental health was being affected. People also voiced their anger and frustration around the idea that mental health was being ignored by the UK Government as a serious issue during the pandemic. Further, there was concern from people about care homes and the isolation being experienced by ‘vulnerable’ elderly residents.  

People also shared their opinion and concern around wearing masks. Some described how they found wearing a mask to be a stressful experience, whilst others were anxious that not everyone would be wearing them. 

A selection of example tweets classified with negative sentiment during this time period is shown in Figure 11. 

Figure11.Exampletweetsfromweeks7to9withnegativesentiment.docx

 

4 Discussion 

In this work, we examined the use of Twitter to track public opinion and concern about mental health during the COVID-19 pandemic. Using a specialist text analytics platform, we collated a large sample of tweets over a nine-week period and carried out various analyses to explore the volume, sentiment, and key topics of discussion.  

The results suggest that this approach is potentially a useful and efficient means to gain a rapid understanding of the key messages, concerns and issues people are facing, at scale. However, at this stage, there are considerable limitations concerning the reliability, accuracy, and transparency of the technology. Qualitative analysis of tweets revealed that some tweets not relevant to mental health had been pulled into the scrape and a significant proportion of tweets came from business or charity organisations rather than from individuals which skewed the sentiment of results. Further, despite the popularity of Twitter, its users are not an accurate representation of the overall demographic of a population.

There is increasing pressure to shape policies and inform decisions at a societal level where robust evidence is lacking. Therefore, future work to further explore the value of this type of work (and this type of data), and any part it might play in informing and underpinning decision making, is recommended  

Acknowledgements 

This project is funded by the National Institute for Health Research (NIHR) [HSRIC-2016-10009/Innovation Observatory]. The views expressed are those of the author(s) and not necessarily those of the NIHR or the Department of Health and Social Care. 

 

References 

1Wang, C., Horby P. W., Hayden F. G., Gao G. F. A novel coronavirus outbreak of global health concern. Lancet. 2020 Feb 15;395(10223):470-3. Available from: 10.1016/s0140-6736(20)30185-9  https://doi.org/10.1016/s0140-6736(20)30185-9  

2Cucinotta, D., Vanelli M. WHO Declares COVID-19 a Pandemic. Acta Biomed. 2020 Mar 19;91(1):157-60. Available from: 10.23750/abm.v91i1.9397  https://doi.org/10.23750/abm.v91i1.9397  

3BBC NEWS. Coronavirus: Two cases confirmed in UK. 2020. Available from: https://www.bbc.co.uk/news/health-51325192 [Accessed 10 November 2020]. 

4Holmes, E. A., O'Connor R. C., Perry V. H., Tracey I., Wessely S., Arseneault L., et al. Multidisciplinary research priorities for the COVID-19 pandemic: a call for action for mental health science. The Lancet Psychiatry. 2020. Available from:   

5Lee, J., Kim J., Hong Y. J., Piao M., Byun A., Song H., et al. Health Information Technology Trends in Social Media: Using Twitter Data. Healthcare informatics research. 2019;25(2):99-105. Available from:   

6Chew, C., Eysenbach G. Pandemics in the age of Twitter: content analysis of Tweets during the 2009 H1N1 outbreak. PloS one. 2010;5(11):e14118. Available from:   

7Signorini, A., Segre A. M., Polgreen P. M. The use of Twitter to track levels of disease activity and public concern in the US during the influenza A H1N1 pandemic. PloS one. 2011;6(5):e19467. Available from:   

8Odlum, M., Yoon S. What can we learn about the Ebola outbreak from tweets? American journal of infection control. 2015;43(6):563-71. Available from:   

9Shin, S.-Y., Seo D.-W., An J., Kwak H., Kim S.-H., Gwack J., et al. High correlation of Middle East respiratory syndrome spread with Google search and Twitter trends in Korea. Scientific reports. 2016;6:32920. Available from:   

10Sinnenberg, L., DiSilvestro C. L., Mancheno C., Dailey K., Tufts C., Buttenheim A. M., et al. Twitter as a potential data source for cardiovascular disease research. JA

11Tavoschi, L., Quattrone F., D’Andrea E., Ducange P., Vabanesi M., Marcelloni F., et al. Twitter as a sentinel tool to monitor public opinion on vaccination: an opinion mining analysis from September 2016 to August 2017 in Italy. Human Vaccines & Immunotherapeutics. 2020;16(5):1062-9. Available from:   

12Bowen, G. A. Document analysis as a qualitative research method. Qualitative research journal. 2009;9(2):27. Available from:   

13Ölcer, S., Yilmaz-Aslan Y., Brzoska P. Lay perspectives on social distancing and other official recommendations and regulations in the time of COVID-19: a qualitative study of social media posts. 2020. Available from: