As the Corona virus propagated very quickly through the world, the topics that make up the key talking points about the Coronavirus changed over time. Wuhan was the initial epicentre of the outbreak. This later became Italy, France, Spain, The United Kingdom and then the United States. Being able to disentangle and track how topics changed over time provides more powerful insight into activity on Twitter and changes in focus as world events change. To do this we use Dynamic Topic Modelling, an extension to Latent Dirichlet Allocation, where documents are grouped into time periods and this time dimension influences topics.

This allowing us to see how topics change overtime. This change consists of differing popularity of a topic over time, but also how a topic shifts over time. For example, the word ‘risk’ in the context of the Coronavirus is not the same as the word ‘risk’ in the context of the gambling. As such, the term usage evolves in a topic over time. It’s also possible that a topic gradually changes it’s meaning over time too as a consequence as word usage changes with the changing local and global situation.

Purpose of Analysis

Dynamic topic modelling is a good tool for data exploration with an obvious temporal dimension. That is, when you don’t know the breadth of content in texts and how they change over time and you want a high level view of what the texts contain and how temporal aspects affect this. With this in mind, our aim is to understand the fundamental topics discussed on twitter in The United Kingdom, France and Italy over time.

** Dynamic topic modelling for France is yet to be conducted **

Caveats and Data Preparation

In preparing the tweet data I took the following steps:

• We kept only: nouns, proper nouns, simplified urls (https://en.wikipedia.org/wiki/Coronavirus_disease -> en.wikipedia.org) and mentions (@SonyCSL). The later two are important in the context of Twitter. Though urls and mentions might not fall within the groups of words one might associate with a topic conventionally, it’s important to remember that a topic in the context of an LDA is simply a group of words that a document with that topic would be more likely to contain.
• Removed infrequent words that are too infrequent to gain adequate statistics about
• Removed very frequent words that occur in many documents and so would simply occur in every topic.

There are several important caveats when considering this data:

• There are days where there was no data or limited data. This undoubtedly has some affect on the resulting topic distributions over time.
• In terms of interpreting topics, there is a large degree of subjectivity involved in recognising and labelling topics. Not all topics have obvious titles and so you shouldn’t take the chosen titles to be definitive. They simply serve to provide a possible simplified interpretation for mental clarity when interpreting the data.
• Just because a topic is not covered, it doesn’t mean it isn’t part of discussion. It just means other topics are better supported by the dataset and modelling limitations.

United Kingdom

Here are the most significant topics from the UK tweets about the coronavirus. For each topic, the most significant terms are shown in subfigures. The titles of topics are manually annotated.

A Closer Look

NHS Staff

Investigating the ‘NHS Staff’ topic we can see that the topic initially starts with terms ‘health’ and ‘system’ alongside other terms. This shifts from the NHS as an institution to care workers and staff over time perhaps indicating the focus on the human effort and sacrifice made by people working for the NHS.

Death Toll

Investigating the ‘Death Toll’ topic more closely, we can see that the focus was on the number of cases per day initially. This then shifted to the number of deaths and was closely tied to Italy for a time when this was the epicentre of the coronavirus outbreak.

Testing

Finally, investigating the ‘Testing’ topic, we can see that there are gradually more mentions of PPE (personal protective equipment) and ventilators over time.

Italy

Here are the most significant topics from Italian tweets about the corona virus. For each topic, the most significant terms are shown beneath. The titles of topics are manually annotated. Not all topics have an obvious title and so don’t take the title as a definitive label. It simply serves to gain a better understanding of the data.

A Closer Look

Countries and institutions

Investigating the ‘Countries and institutions’ topic we can see that the topic initially starts with a focus on the usa, europe and the economy. These are likely general topics discussed in Italy before the outbreak. This topic becomes less about the USA and the economy over time, focusing more on Europe and the European Stability Mechanism (Meccanismo europeo di stabilità) and European Central Bank (Banca Centrale Europea). It is clear that discussion about how the European Union will respond to the crisis started to become a dominant topic.

Italy & Measures

Investigating the ‘Italy & Measures’ topic more closely, we can see that the overall focus is Italy. Other than this, the topic changes a lot over time.

• We see quarantena (quarantine/lockdown) is initially high (probably due to discussion about Wuhan) and reduced slightly before peaking around the time Italy went into it’s own lockdown.
• Control (controlo) remains a fairly important topic over time.
• Volo (flights) are discussed, likely related to cancelling flights or spread from flights.
• The topic starts off with nave (boat) as a talking point related to immigration. This is likely mixed in with words like controlo (control) initially but that context drops away over time.
• There is a lot of discussion about borders (bordare) until the end of February when this topic becomes less important.

Spread & Measures

Investigating the ‘Spread & Measures’ topic more closely, we can see:

• ‘Rischiare’ (risk) becomes the dominant word of the topic until mid February. Likely representing the increased attention of the risk of coronavirus over time.
• ‘Rischiare’ increases with ‘diffusione’ identically at the start, perhaps indicating these concepts are closely linked initially. They remain important but vary, as indeed ‘risciare’ in the context of the coronavirus covers many potential risks while diffusion is a technical term concerning the spread of the disease.
• ‘Misura (measure) becomes the most significant word of the topic.
• We can see that ‘allarismo’ (alarm) and ‘quarantena’ (quarantine) are initially part of the topic but the focus seems to be more on spread of the disease rather than lock down measures
• ‘Lockdown’ becomes the most important word in this topic at the end, growing from the start of march.

Regions

Finally, investigating the ‘Regions’ topic. We can observe that:

• Initially ‘Hubei’ and ‘province’ are the focus.
• The focus then shifts ‘lombardia’ (a badly affected area in Italy) and ‘regione’ (region).
• We also see Veneto (another region) affected at roughly the same time as Lombardie.
• Finally we see Bergamo is discussed, another hotspot for coronavirus in Italy.

Conclusions

Dynamic topic modelling has allowed us to identify how topics and topic word composition change over time as the local and global situation of the Coronavirus changed. Sometimes the shift of topic dominance and topic term composition is quite rapid as the global context changed. The most obvious case of this is the changing epicentres of the disease from Wuhan to Lombardy, Italy as a whole and France, the United Kingdom and Spain.

Italy and the United Kingdom have many similar topics surrounding the Coronavirus. However, there are differences. Perhaps the most obvious differences are:

• Italy’s discussion of the role the European Union has to play contrary to the UK that doesn’t discuss the EU outside of the Brexit that surprisingly does not seem to take centre stage in topics very much either.
• Italy’s initial focus on African Immigration while immigration doesn’t seem to be a major topic in the UK.
• Italy’s focus on particular local regions such as Lombardy, Veneto and Bergamo while the UK doesn’t seem to focus on particular regions.