COVID‐19 Tweet Links: A Preliminary Investigation of Type and Relevance

ABSTRACT

We conducted an exploratory study of the links found in Twitter tweets. Our results showed that the largest category of tweet links was social media platforms followed by alternative news sites. Government agencies and educational institutions were under‐represented. In terms of relevance, about 75% of the links were related to COVID‐19 but disappointingly, only 40% of the links were directly related to their respective tweets' topics.

INTRODUCTION

During the COVID‐19 pandemic, users have generated an immense quantity of social media content. Twitter tweets, in particular, have often been used for analysis because they are regarded as a valuable source of data in the surveillance of diseases (e.g. Sinnenberg, et al., 2017). More importantly, Twitter is in active use by governments, organizations and individuals for the sharing of COVID‐19 information.

The restricted length of a tweet is a double‐edged sword. On the one hand, information is packaged into small chunks that can be quickly read. On the other, there is a limit to the amount of information each tweet can convey. Consequently, links are typically embedded, presumably to lead users to Websites with elaborated content or as a citation source. However, it is evident in many contexts that not all links serve these purposes. For example, links may be clickbait or serve other malicious purposes (Jamison, Broniatowski, & Quinn, 2019).

In the COVID‐19 context, tweet links play an important role in the dissemination of public health information. However, if misused, fake news and other forms of disinformation may be spread instead. Surprisingly, there is, to our knowledge, little current work that has investigated the types and relevance of tweet links found about COVID‐19. The present study thus aims to answer the following questions: (1) What types of Websites do the tweet links about COVID‐19 point to? (2) Are these Websites relevant to COVID‐19 and their respective tweet topics?

METHODOLOGY

Data were drawn from a project that collected COVID‐19‐related tweets (Chen, Lerman, & Ferrara, 2020). We selected a date range of one week, between 9 to November 16, 2020, comprising 189,071 tweets. This period was chosen as it represented one week after Pfizer and BioNTech announced the first vaccine efficacy results (Business Wire, 2020), providing hope, controversy, and plenty of online chatter. Next, tweets without links were discarded, resulting in a remainder of 56,577. Following this, a random sample of 4,001 tweets was selected. This random sampling approach is consistent with prior public health research using tweets such as Cavazos‐Rehg et al. (2016). Finally, for each of the 4,001 tweets, its link was followed by three researchers independently. The link was checked to see if it was still accessible. If so, the type of Website was noted. Next, the link’s content (Website) was read and ascertained if it was relevant to the tweet’s topic. Once all the tweets were analyzed, the researchers discussed their respective results to resolve differences until consensus was reached.

RESULTS

Our analysis showed that the largest category of tweet links was social media platforms such as Facebook, accounting for 2032 (51%) links. Next came alternative news sites at 614 (15%) links, followed by mainstream news sites at 531 links (13%). At the other end of the spectrum, there were only 62 government Websites (1.5%), with online stores and educational institutions having the smallest share at 27 (0.7%) links. Approximately 7% or 285 links were not accessible. Table 1 shows the distribution of links by type of Website.

TABLE 1.

Distribution of links by type of Website

	All Tweets		Relevant to COVID‐19		Relevant to Tweet
Type	Number	Percentage	Number	Percentage	Number	Percentage
Social media	2032	50.79	1,597	78.59	760	37.40
Alternative news	614	15.35	554	90.23	353	57.49
Mainstream news	531	13.27	506	95.29	309	58.19
Commercial	113	2.82	72	63.72	31	27.43
Government	62	1.55	52	83.87	36	58.06
Non‐profit	52	1.30	40	76.92	18	34.62
Databases	43	1.07	34	79.07	10	23.26
Personal website	30	0.75	24	80.00	17	56.67
Online store	27	0.67	7	25.93	4	14.81
Educational institution	27	0.67	24	88.89	16	59.86
Other	185	4.62	100	54.05	43	23.24
Not accessible	285	7.12	‐	‐	‐	‐
Total	4,001	100	3,010	75.23	1,597	39.92

To answer the second question on relevance, the final row of Table 1 shows that of the 4,001 links in our dataset, 3,010 (75%) were relevant to COVID‐19. However, there were only 1,597 (40%) links that were directly relevant to their respective tweets' topics, meaning that the majority of links were not topically relevant.

Delving deeper into each category, the top three link categories that were most relevant to COVID‐19 were mainstream news, alternative news and educational institution Websites. The three least relevant categories were commercial Websites, online stores, and those that were not classifiable. Link categories that were most relevant to their respective tweets' topics were educational institutions, mainstream news and government‐related Websites. Conversely, those that were not relevant were online stores, databases and those that were not classifiable.

DISCUSSION AND CONCLUSION.

Our findings show that while there was a variety of link types posted in our dataset of COVID‐19 tweets, their distribution was uneven. In particular, we were surprised to see that slightly more than half of the links pointed to social media content. Unfortunately, sites where credible information could presumably be found, such as mainstream news, governmental organizations and educational institutions were not well‐represented. Collectively, these three categories comprised only about 15% of all links analyzed. Our results reinforce the prevailing notion that social media is a major source of news.

While it was heartening to see that the majority of tweet links (approximately 75%) were relevant to COVID‐19, this also meant that around a quarter were unrelated, which is not an insignificant number. Alarmingly, when we checked if the links were both relevant to COVID‐19 and its respective tweet topic, slightly less than 40% met this criteria. Taken together, the under‐representation of certain categories of information sources together with the large proportion of non‐relevant links may mean that people who use social media as a platform for information seeking may at best, not meet their information needs effectively, and at worst, fall prey to misinformation.

One important implication from the user's perspective is the need to be vigilant about links in online content, such as tweets. Users may exploit a hot topic, such as COVID‐19, to advance an agenda, whether it is spam, advertising, or other deleterious aims. Admittedly, this may be challenging as URL shorteners are often used, masking the actual Website being referenced. The relative lack of links from official or more credible sources such as government agencies, educational institutions and mainstream news is another concern. Possible reasons include a lack of interest in posting such links by users, low levels of engagement by these organizations, or there are simply more people keen on posting other types of links. This requires further investigation.

There are limitations in the present study that warrant future work. Due to the large volume of data. we did not investigate the quality of the Websites being linked. As well, we were not able to analyze all the tweets during the study's date range. Additionally, more recent time periods may yield different outcomes. Future work may also consider analyzing the links of other social media platforms.