Public engagement with science on social media: Twitter vs. weibo

Social media have become an integral part of the public engagement of science and technology, enabling participatory and reciprocal conversations and interactions between specialists and audiences [12, 13]. Specialists, including scientific communities, ethicists, and policymakers, increasingly use social media to open in-depth discussions about revolutionary technologies with the general public [14–16]. Social media have changed how the general public seeks information about societal issues and how they learn about the latest scientific advances [6]. Through social media, users express their engagement mainly through “content interactivity,” how users “control the information they receive” [17], and “human interactivity,” how users comment on content and share content with others [18]. The content on social media has been studied to examine different scientific issues, for instance, climate change [19] and vaccination [20, 21]. These studies have shown that social media have become important communication tools for scientific issues.

As a social media platform, Twitter serves as an important communication channel for the public engagement of science in the U.S. and worldwide. The majority of users are from the United States, with more than 47 million accounts [22]. For communicating science to broader lay audiences, Facebook and Twitter are of particular interest in the U.S. Although Facebook is the most popular social networking site [23], its platform features, such as restrictions on user activity to only accepted “friends,” make it difficult for researchers to connect with the public [24]. Unlike Facebook, tweets from unprotected accounts can be seen by everyone [25]. Further, Twitter discussions and real-time updates have the potential for the public engagement of science [24]. Consequently, numerous scientists from many disciplines use Twitter for public engagement [16].

Weibo, the equivalent of Twitter in China, is one of the largest social networking sites. Despite being the most popular microblogging service in China, there is limited research on how Weibo is employed for public engagement with science. For example, Yu et al. discovered that Weibo is an important local altmetrics platform on which global science is discussed [10]. Further, Rauchfleisch and Schäfer discussed how Weibo facilitated the information flow of sensitive topics, such as food safety and climate change [26].

Twitter and Weibo are similar in many aspects. Both are recognized as the top microblogging services with hundreds of millions of users worldwide [27]. The social networking elements are similar, and both enable users to post, repost, or forward messages [28]. Like Twitter, Weibo is also a directed social network, allowing users to follow other users and browse real-time trending topics. Both platforms heavily rely on advertising revenue [29].

Although Weibo has several similarities with Twitter, the social media platforms have different characteristics, which may influence public engagement with science. Twitter’s interface is less media-rich compared to Weibo. Twitter messages are limited to 280 characters and users can upload photos or short videos, while Weibo lets users insert animated emoticons, images, videos, music, and even polls. Furthermore, the types of trending topics on the two platforms are different. The trending topics on Weibo are often about entertainment and rarely about politics or science. While the trending topics on Twitter are more diverse; there are more discussions regarding science and politics [30, 31]. In addition, users’ demographics differ between the two platforms. U.S. Twitter users are about 40 years of age on average, highly educated, wealthier than the general public, and more likely to identify as a Democrat than a Republican [32]. However, 40% of Weibo users are 23 to 30 years old, and about 35% are aged 18 to 22 years [33].

Scientific literacy and public discussion on social media

Social media, especially microblogging sites, have gained considerable academic attention for facilitating public discussion due to its open access and connected structure [34]. Social media have provided a space where privileged sectors and ordinary citizens, including small and marginalized groups, can create a public presence and discuss public issues. Public discussions cover a wide variety of topics, including family life, business, science and technology, and health [5, 35]. While people in different parts of the world may discuss the same issue, discussions may differ across countries [36].

Scientific literacy levels of engaged users may influence the public discussions of cutting-edge science issues on social media. Scientific literacy is considered one of the critical competencies for social inclusion, which will help any citizen access, read, and understand the world with a scientific dimension [37]. Various academic scholars argue that having a scientifically-informed population is essential for democratic processes in a technologically demanding society [38]. High literacy levels can lead to more reasoned and rational public discussions, which is vital for the informed decision-making process of legislators and industry [39]. These differed between the US and China [40]; for example, scientific literacy levels among Chinese citizens were relatively low compared to the U.S. and European countries in 1995 and 2005 [40]. Because of these differences, voices regarding gene-edited babies on Twitter could likely be distinct from those on Weibo.

Further, understanding perceptions of gene editing from a cross-country perspective allows researchers to examine how social media may shape the landscape of user opinions on science topics [6]. Since individuals from different cultural groups may receive and discuss different information online, they may have distinct resulting attitudes and opinions. Because social media is pervasive in our everyday lives, we can examine the primary topics that emerge from each platform and how they may shape or reinforce the platform users’ beliefs and opinions of gene editing. This understanding will help guide future message strategies for science communication.

Discussions about gene editing

Human application of CRISPR have raised profound concerns, challenging scientific communities, bioethics groups, policymakers, and the public to think about the existing policy for governing human genome editing [41]. There are some concerns that gene editing science and innovation are moving ahead of public understanding and legislation [42]. It is urgent to call for public engagement and a broad public discussion about the ethical, legal, and wider societal implications of this technology.

Mainstream discussions regarding gene editing are usually about ethical and social concerns, including the safety of the technology, the alteration of the human genome, issues of eugenics, and regulation policies [43–45]. Though CRISPR has increased further specificity and reduced off-target effects, the risk of inaccurate editing is still a concern. He Jiankui has attempted to modify the CCR5 gene in embryos. Thus, there could be possibility that off-target changes in other genes were made in the twins’ genome [46]. Besides, once the mutations are introduced into the human population, the altered genes would be difficult to remove and may spread across communities or even countries [47]. The societal risks are as important as safety risks. Genetic enhancement of humans could result in social inequality and discrimination in society. Multiple polls indciate that there is greater support for therapy (disease treatment) than enhancement (e.g., improving memory and learning capacity) across different countries [48–51]. Many people oppose genetic enhancement in humans as it is intrinsically eugenic in nature; the inheritable genetic enhancement for “more desirable” and “better” kinds of humans [52].

Discussions on social media may influence the public’s knowledge and attitudes toward CRISPR/Cas9. Media have the power to construct public beliefs and attitudes about topics and events, especially when individuals have little prior knowledge of these subjects [53]. Since previous research has established the relationship between gene editing knowledge and science attitudes [54], examining the discussion of gene editing on social media is particularly useful. Several recent studies have provided some insights into online discussion about gene editing. For example, Calabrese et al. examined the discussions about CRISPR/Cas9 on Twitter in 2018, uncovering four main themes, including research and applications of CRISPR, the breaking news story of the use of gene editing on human embryos in China, biotechnology regulations in agriculture, and research relating to muscular dystrophy [5]. Further, the authors found that the “CRISPR babies” news story dominated discussions of gene technology, indicating that major events impact how users discuss socio-scientific issues in online media [5]. Walker and Malson undertook a textual analysis of Facebook comments on posts by news media outlets about agricultural and environmental gene editing and found four themes emerged, including gene editing as challenging a higher power, the conflation of gene editing with genetically modified organisms (GMOs), pro-science arguments, and comparisons to science fiction [55].

Because social media can shape people’s beliefs and attitudes toward new topics like gene editing, it is important to understand what users discuss on their respective platforms. Further, understanding the differences and similarities of how Chinese and American social media platforms discuss gene-edited babies, likely due to cultural differences, may promote mutual understanding and facilitate informed policy decisions. Moreover, discussions about scientific issues could dramatically change over time, especially during large events like the announcement of the first gene-edited babies. Examining the evolution of the topic distributions around the gene-edited babies event will help policymakers and science communication practitioners develop more effective communication strategies with the public. Thus, we propose the following research questions:

1. R1: What topics are discussed about CRISPR and gene editing on Weibo and Twitter?
2. R2: What are the differences of discussed topics between Weibo and Twitter?
3. R3: How do the discussed topics evolve over time?

Data collection

The terms “gene editing,” “genome editing,” “CRISPR,” and “CRISPR-Cas9” have been shown to be salient keywords on Twitter [5]. We conducted a preliminary search on Weibo using the corresponding Chinese keywords, and the majority of the obtained posts were closely related to gene editing technology. In addition to the above four terms, “genome engineering” is also widely used as “gene editing” and “genome editing” on Chinese social media. Therefore, to make the dataset comparable, we used “gene editing,” “genome editing,” “genome engineering,” “CRISPR,” and “CRISPR-Cas9” as searching keywords for Twitter, and “基因编辑,” “基因组编辑,” “基因组工程,” “CRISPR,” “CRISPR-Cas9” for Weibo. The search period was set two months before and after the Chinese gene-edited babies event to fully understand the public discussions, ranging from September 25, 2018, to January 25, 2019. Weibo and Twitter users have generated far more tweets on the gene-edited babies event during this period than ever before. We collected 11,244 posts from Weibo through Zhiwei, a well-known social media data provider, which collected post based on Weibo API, and 57,525 tweets from Twitter based on its Premium Search Tweets API. Weibo may have filtered some posts in compliance with the rules and laws of the Chinese government. Also, due to new Weibo functions that allow users to change their settings so that others only see posts within the most recent six months, some posts could not be collected.

Data pre-processing

Topic modeling

Latent Dirichlet allocation (LDA) [58, 59], a widely used tool to extract topics from text data, was applied to Twitter tweets and Weibo posts. LDA is an unsupervised probabilistic model that generates mixtures of latent topics from a collection of documents, where each topic is characterized by a distribution over words. A distribution over topics is first sampled from a Dirichlet distribution, and a topic is chosen based on this distribution. LDA is a mathematical method for estimating both of these simultaneously: finding the mixture of words associated with each topic while also determining the mixture of topics that describes each document. Each document is modeled as a distribution over topics, with topics represented as distributions over words [60]. To find these topics, LDA uses word co-occurrence patterns in the corpus, such that the more often two words co-occur in a document, the more likely they are to be assigned to the same topic [61]. An important task for topic modeling is to determine the prior value for the number of topics (k); if k is too small, the topics will be overly broad, while if it is too large, the topics may be too overlapping or similar to each other.

To efficiently run topic modeling for larger datasets, we used LDA in this study. For tweets, the mean length is 119, and the median length is 131. For Weibo posts, the mean length is 116, and the median length is 142. The distributions for document lengths of tweets and Weibo posts are shown in S3 Fig in S1 File. Moreover, for topic quality, LDA can achieve a similar performance as other short text topic models like GSDMM and Biterm [62–64].

It is crucial to understand tweets and Weibo posts accurately. Hence, we applied two different topic models: one for tweets and one for Weibo posts. This has several methodological advatanges. First, due to the need to compare two different languages, English and Chinese, applying separate models removes issues relating to translating documents or the accuracy of the translations. Translation services are often not satisfactory for delivering accurate and context-specific translations, which may cause confusion. Because the U.S. and Chinese researchers in our team can fully understand and interpret the contents, utilizing two separate models is ideal. Second, applying a single cross-collection topic modeling, rather than two separate models, might allow some topics specific to one dataset to be overshadowed and ignored. For example, many of the topics on Weibo were specific to China and Chinese media, which may have been ignored with the addition of Twitter discussions. Third, we wanted to conduct a qualitative analysis of the differences between the U.S. and China and explain the differences from the communication perspective.

Following the steps recommended by [59, 65], we built and validated the models based on the following procedures. First, we selected the models based on the topic coherence score, which measures a topic’s semantic interpretability and association with well-defined semantic concepts [66]. As a high score denotes meaningful and interpretable topics, we trained multiple LDA models and selected models with relatively high coherence values at peaks (Fig 1). For Weibo posts, we chose the models when k was 3, 7, 9, 11, 13, and 14; For tweets, we selected the models when k was 5, 7, 9, 11, 12, 13, and 14. Second, researchers interpreted the topics based on the most frequent keywords and a sample of documents with the highest probability. During the process of interpreting, we also used intertopic distance mapping to assist with the determination of topic selection. The distance map was drawn and visualized using the Python package pyLDAvis [67], where the distances between topics are first calculated using Jensen-Shannon divergence, and then the intertopic distances are projected onto two dimensions [68]. If they were overlapping, the semantic contents of the topics would be similar. Based on interpretation and intertopic distance map, we chose models that were well interpretable and not overlapping, as shown in S4 Fig in S1 File. Finally, we re-evaluated the validity of interpretation by checking a sample of automatically assigned documents. S1 and S2 Tables in S1 File list the reasons why we did not accept other models. Therefore, we set the number of topics as 3 for Weibo posts and 9 for Twitter. The example tweets for 9 topics and example Weibo posts for 3 topics are listed in S3 and S4 Tables in S1 File.

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Fig 1. Coherence scores for weibo posts and tweets.

https://doi.org/10.1371/journal.pone.0267406.g001

Topics on weibo

As is shown in Table 1, there were three topics about gene editing during this period on Weibo. Topic 1 talks about the Chinese government’s investigation into Dr. He Jiankui, the Associate Professor from Southern University of Science and Technology (SUST) (Guangdong is the province that includes the city of Shenzhen, where SUST and the Hemei Hospital that allegedly gave ethics permission for the experiment is located). Chinese authorities from the Guangdong province placed Dr. He under investigation and claimed that his work defied government bans and was “in the pursuit of personal fame and profit.” However, they also found that Dr. He “deliberately evaded supervision, raised funds and organized researchers on his own to carry out the human embryo gene-editing intended for reproduction, which is explicitly banned by relevant regulations [69].” This topic accounted for 21.8% of the tokens.

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Table 1. Topics for Weibo posts based on LDA.

https://doi.org/10.1371/journal.pone.0267406.t001

Topic 2 concerns scientific aspects of CRISPR technology, especially the achievements based on the use of the technology. One of the most popular achievements discussed on Weibo was on January 24, 2019, where Chinese scientists cloned five monkeys from a gene-edited macaque by somatic cell nucleus transfer (SCNT). Clones produced this way may be useful for the development of non-human primate models of human diseases (circadian rhythm disorders). Another news story is from XinhuaNet where a U.S. genetics company announced the launch of a clinical trial using CRISPR to treat hereditary eye disease. Topic 2 involved a third (33.9%) of the tokens.

Topic 3 concerns the International Summit on Human Genome Editing controversy at the University of Hong Kong. It involved the most tokens, 44.3%. Dr. He was invited to speak at the Summit and described how his research team had conducted experiments on cultured human embryos to disable a gene known as the CCR5 gene in an attempt to confer genetic resistance to human immunodeficiency virus (HIV). According to the percentage of the tokens, most of the posts on Weibo are about the Summit in Hong Kong.

Topics on twitter

Nine topics about gene editing were discussed on Twitter, as shown in Table 2. It also lists the percentage of tokens for each topic. Topic 1 depicts the commercial aspects of CRISPR technology, such as the market value of CRISPR technology, the market share of the biotech companies (CRISPR Therapeutics AG), and the CRISPR/Cas9 intellectual property battle. Topic 2 is about biohacking CRISPR technology, by one famous biohacker, Josiah Zayner, who runs a company called the Odin out of his garage in Oakland, California, selling biohacking supplies, including DIY kit. He injected himself with DNA from CRISPR at a biotech conference. Topic 3 emphasizes that CRISPR is a powerful research tool, especially its targeting specifics. These tweets are usually about scientific research from academic journals, such as Nature Biotechnology, bioRxiv.org, and Nature Cell Biology.

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Table 2. Topics for tweets based LDA.

https://doi.org/10.1371/journal.pone.0267406.t002

Topic 4 involves science communication about CRISPR technology. Many tweets share a link to a popular video on YouTube named Genetic Engineering Will Change Everything Forever, which introduces CRISPR. Also, a large number of tweets linked to an article from Vox titled “A simple guide to CRISPR, one of the biggest science stories of the decade.” Topic 5 discusses the consequences of conducting controversial experiments. For example, one tweet stated, “CRISPR-baby scientist fired by the university.” Geneticist Robin Lovell-Badge of the Francis Crick Institute in London was worried that Dr. He could face the death penalty [70]. Topic 6 focuses on the advantages of CRISPR technology, and mainly relates to two news articles. One article is titled, “Antibiotics Are Failing Us. Crispr Is Our Glimmer of Hope #AntibioticResistance”, and the other is titled, “CRISPR Repurposed to Develop Better Antibiotics.” According to Rettedal (2019), “antibiotics are still massively overprescribed, a new study shows. With no new drugs in sight, some scientists are turning to CRISPR for a reboot” [71].

Topic 7 strengthens the development of genetically modified spicy tomatoes. Despite this controversy surrounding He Jiankui, scientists were still interested in CRISPR and might soon create spicy tomatoes by switching on their chili genes, according to MIT Technology Review. Topic 8 describes the clinical application of CRISPR, especially its use for the treatment of brain cancer and for medicine. Topic 9 has the largest proportion of the tokens and focuses on the ethics of the human application of CRISPR, including the world’s first gene-edited babies. This is most likely due to the period during which the posts were collected.

Topic distributions over time

Fig 3 depicts the evolution of the topic distributions over time on Weibo and Twitter. Dr. He announced the birth of twin girls with edited genomes on November 25, 2018. The topics peaked on November 27, 2018 (Beijing time) on Weibo and peaked on November 26, 2018 (New York time) on Twitter. The discussion of these topics then decreased rapidly, indicating the discussion of CRISPR technology was primarily driven by the designer babies event. On Weibo, most of the posts were about the gene-edited baby controversy discussed at the Summit, and the least was about the investigation into Dr. He’s research on November 26, 27, and 28, 2018. Then, gene editing quickly faded from public attention, and only a few posts mentioned the scientific aspects of CRISPR technology. However, the public discussed the investigation into Dr. He’s research again in January 2019. On Twitter, people seldom discussed gene editing before November 25; however, all the topics peaked on November 26. Generally, more attention was paid to gene editing after the event. The most discussed topic was the ethics of designer babies, which peaked on November 26, 2018. Another highly discussed topic was communicating the science of CRISPR technology. The event of gene-edited babies by Dr. He may have stimulated public interest in gene editing, and many people may be starting to learn about this technology.

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Fig 3. Temporal distribution of topics on Weibo (A) and Twitter (B).

https://doi.org/10.1371/journal.pone.0267406.g003