Microblogging and nanotweets: Nanotechnology on Twitter

Abstract

The social web represents a new arena for local, national and global conversations and will play an increasing role in the public understanding of science. This paper presents an analysis of the representations of nanotechnology on Twitter, analysing over 24,000 tweets in terms of web metrics, latent semantic and sentiment analysis. Results indicate that most active users on nanotechnology are distributed according to a power law distribution and that web metric indicators suggest little conversation on the topic. In terms of content, there is a remarkable similarity with previous studies of nanotechnology’s representations in other media outlets. Related to content is the sentiment analysis that indicates predominantly positively loaded words in the corpus. Negative sentiments mainly took the form of uncertainty and fear of the unknown rather than open hostility.

Keywords

latent semantic analysis nanotechnology opinion mining social web Twitter

1. Introduction

The portrayal of scientific and technological objects in the mass media plays a fundamental role in our understanding of how such issues are discussed and interpreted in the public sphere and constitute common sense knowledge (Moscovici, 2000). There is a wide scientific literature on this topic that combines communication theories and studies of mass media influences (e.g. agenda setting theory, McCombs and Shaw, 1993; second level agenda setting, McCombs and Ghanem, 2001; cultivation theory, Gebner and Gross, 1976; agenda building, Lang and Lang, 1983; second level agenda building, Fahmy et al., 2011) with public understanding of science research. In particular, studying how mass media frame an emerging technology is important for observing definitions and associated meanings that are legitimised or stigmatised. Science in the media can be described in terms of both the kind of medium analysed (e.g. printed media, television, movies, Internet and the World Wide Web) and the theoretical approach to the study of the role of media in the public perception of a new technology. Clearly, the choice of the media analysed also influences the theoretical and methodological approaches. For example, years after the advent of television as the main mass medium, printed media were used comparatively more frequently because textual data could be analysed in both quantitative and qualitative approaches following consolidated methodologies. This is an exploratory study of the use of the social web platform Twitter¹ to share and discuss news on a “new” technology: nanotechnology. It aims to contribute to the literature as one of the first studies on the use of the social web in the context of the public understanding of science.

The earliest studies on science in the media obviously focussed on printed media. The focus was on non-fiction scientific journalism and it produced a great deal of research about science journalists and the “quality and quantity” of their media output. Many studies focussed on “objective” measures of media content, such as accuracy and readability (see e.g. Dunwoody and Scott, 1982; Bader, 1990). These studies reinforced the idea that “simplifications” had been made in the way science was covered and implied that this was the reason for the low level of accurate scientific knowledge among the lay public, and also for negative attitudes towards science in general. Since many of these studies were based on a linear communication diffusion model (Bucchi, 2002) of scientific information, most of them concluded that not enough information was published and that what was published was not provided in sufficient quantity or detail to have been useful for the public in forming an opinion about the scientific or technological fact discussed. The result of this approach to the study of the role of mass media in science communication was a focus upon the accuracy of the scientific information published. Hence, the mass media was considered a “dirty mirror” (Friedman, 1989) of science, or sometimes even as the direct cause of a hostile public reception as the result of the negative coverage of a given scientific issue.

However, the deficit model approach and linear model of communication were criticised for a simplistic view of the processes of human communication and interpretation. Several studies argued that media coverage of controversies cannot simply be “improved” by better “dissemination” of scientific or technological information because it is shaped by structural relationships within communities and cultural contexts (Wilkins and Patterson, 1991).

Other studies included the cultural contexts in which the framing of a new technology took place in the mass media and discussed the communication processes involved. Some examples of this approach are: Gamson and Modigliani (1989) on nuclear power and cultural resonance of symbolic resources; Nelkin (1994) on the use of metaphors by scientists; and Neidhardt (1993), who looked at the public as a communication system.

A different approach was taken in one of the studies based on social representations theory (Moscovici, 2000), in which the aim was to identify representations of a given scientific or technological issue, their adoption by social groups and their role in the formation of common sense knowledge. Communication plays a crucial role in social representations formation. Studies in this area are focussed on the idea that in the public sphere there are competing definitions, in what is a complex game played for the control of semantics in the public sphere (Gaskell, Bauer and Durant, 1998). Considering that definitions are not just technical issues, but are a matter of framing for the purpose of opinion and attitude formation and for regulation, competing representations in the media is a field where the battle “is being waged in the arena of language, as much as that of science” (Ogden, 2001: 340). The social representations theoretical framework has informed several studies on media representations of scientific issues through the notion of “anchoring” using both qualitative and quantitative methods or a combination of both (Veltri and Suerdem, 2011; Veltri, 2012).

2. Other media and the Internet

In parallel to the theoretical shift from a linear diffusion model to a more interpretative approach, after the analysis of printed media, a new emphasis was placed on other mass media, mostly on television, as well as film, novels, etc. (e.g. Gopfert, 1996; Jones, 1997; Holliman, 2004). Researchers who analysed television news/documentary shows and their representations of scientific and technological issues faced the difficulties in defining the sampling techniques and the absence of a largely shared and approved analytical method that could be applied to different types of data. Yet, there is a wide range of studies on the televisual framing of a scientific or technological issue, either using adapted content and thematic analysis from printed media (e.g. Leon, 2008) or adopting more interpretative approaches (e.g. Hoijer, 2010).

Another important change occurred in the media with the advent of the Internet and the World Wide Web; both attracted the attention of scholars aiming to study the role of websites in the dissemination of scientific knowledge (e.g. Eveland and Dunwoody, 1998; Byrne et al., 2002). Initially, much emphasis was placed on a linear and mono-directional dissemination of information that conceptually resembled the way the “old media” worked, but later the interactive and social features of Web 2.0 were taken into consideration (Waldrop, 2008).

Recently, online media have started to be considered for their potential as public arenas for discussion among individuals in the forms of mailing lists, chat rooms and discussion forums (Rogers and Marres, 2000; Triunfol, 2004; Delborne et al., 2011). This goes back to the theoretical approach to interpersonal conversation about science and technology and the processes of negotiations of meanings.

The incredibly fast diffusion of social web services complicates the roles that the social web plays in the ways individuals seek and process information in contemporary society. As virtually costless and extremely versatile channels of diffusion, as tools for coordinating and mobilising social movements and their capacity for “crowd-sourcing” (Shirky, 2010) large cognitive tasks are just some examples of characteristics of these new media. Social web platforms like Facebook and Twitter are often considered as “conversational hubs”. In the case of Twitter, Zappavigna (2011) argues that there is a shift to a “searchable” discussion.

Similarly, there have been an increasing number of studies using Facebook as a source of data and as a medium to study the sharing of information other than that of social interaction. Lewis et al. (2008) discussed the potential use of Facebook data in the context of merging “virtual” and “real life” variables and the identification of specific network behaviours. Examples of this are more and more frequent, for example Waters et al.’s (2009) study on non-profit organisations and their use of Facebook, or Westling (2007) on the impact upon political communication and campaigns. In addition, there are increasing numbers of studies on the role of the social web in the diffusion and understanding of information related to health (Hawn, 2009; Jain, 2009).

There is also evidence of an increasing role of the social web in the information foraging behaviour regarding news online. Social networks are becoming increasingly more popular as venues for sharing content with friends than are blogs, email or message boards, according to a report by AOL and Nielsen Online (Pew Research Center, 2010). The analysis of 10,000 social media messages found that 92% of Web users used social sites to share content with their friends, with 60% of the material shared having links to published content such as news and media sites. With the rise of the participatory web and social media and the resulting proliferation of user-generated content, the public potentially plays a far larger role in all stages of knowledge translation, including information generation, filtering and amplification.

Given social media’s dual nature of information source and conversation enabler, this study explores the role of Twitter in the sharing and discussion of information on nanotechnology, bridging the relevant theoretical and methodological literature from new media studies. According to Marwick and Boyd (2011), in the context of posting information about social issues the main reasons for re-tweeting are to spread or emphasise tweets to new audiences, to comment on someone else’s tweet by re-tweeting and adding new content and to publicly agree with someone.

Social media and in particular Twitter are used as a source of data as new “infoveillance” methods such as mining, aggregating and analysing online textual data in real time are becoming available (Pang and Lee, 2008). Twitter is potentially suitable for longitudinal text mining and analysis. The brief (≤140 characters) text status updates (“tweets”) users share with “followers” (e.g. thoughts, feelings, activities, opinions) contain a wealth of data. Mining these data provides an instantaneous snapshot of the public’s opinions and behavioural responses. Longitudinal tracking allows identification of changes in opinions or responses. In addition to quantitative analysis, the method also permits qualitative exploration of likely reasons why sudden changes have occurred (e.g. a widely read news report) and may indicate what is holding the public’s attention.

An example of applying this methodology to scientific issues is Chew and Eysenbach (2010) in which the authors monitored the spread of information on H1N1 (or “swine flu”) in Twitter using it as a real-time tracking tool of public attention on the pandemic. However, there have been very few studies analysing the use of Twitter to represent a new technology or to “anchor” it.

3. Representations of nanotechnology

Studies of nanotechnology in the mass media have mainly analysed printed media and there are few studies about this new technology’s representations in other media. Lewenstein, Gorss and Radin (2005) tracked nanotechnology in the American press from 1986 to 2004. According to the authors, media attention to nanotechnology seems to mimic the coverage of biotechnology in its early stages as a public issue: it started with low levels of attention, which then rose sharply as the issue spread from elite media to general outlets. As with biotechnology, the initial coverage of nanotechnology was largely positive and focussed on progress or the potential economic benefits and with relatively little discussion of risk. Similarly, Scheufele and Lewenstein (2005) also made predictions about key themes of coverage: foci on the economic potential or the scientific promise of nanotech early in the issue cycle. Friedman and Egolf (2005) analysed the US and UK coverage of nanotechnology’s risks between 2000 and 2004 using the most prominent national newspapers and newswire services and concluded that the UK had a slightly more negative coverage than the US and placed a larger emphasis on societal risks. This is in line with the findings of Gaskell et al. (2005) that compared public attitudes towards nanotechnology in the UK and US and also provided an analysis of nanotechnology’s coverage in one British and one American national newspaper (The Independent and The New York Times). Their research showed that the American newspaper covered the potential benefits of nanotechnology more than the British one; it also showed that coverage of risks appeared later compared to that of benefits.

Although Europe, or at least the UK, produced more negative coverage than the US, this does not mean that it was overall negative. Anderson et al. (2005) analysed the framing of nanotechnology in the British press, and suggested a considerable amount of uncertainty about the nature of nanotechnology with a tendency for being optimistic and highlighting the potential benefits more than the risks.

In the few other European countries studied, nanotechnology coverage has been overall positive indeed, for example in Norway (Kjølberg, 2009) and Denmark (Kjærgaard, 2010). The main themes that emerged in these two countries were: “nanotechnology as important for the future”, “nanotechnology is under control” and “benefits outweigh risks”.

In the German national press, nanotechnology also appears to be predominantly framed in positive terms with an emphasis on medical and economic benefits (Donk et al., 2012), very much in line with the above described studies. A recent study on nanotechnology’s coverage in Slovenia (Groboljsek and Mali, 2012) highlights a similar pattern although there was very little coverage by local media.

Outside the analysis of newspapers’ coverage, Landau et al. (2008) studied the impact of visual images of nanotechnology, identifying a specific visual domain of science images that are organised according to polarities affecting the volatility of attitudes towards the technology under scrutiny. Similarly, Hanson (2012) analysed the evolution of nanotechnology’s images and their tropes which relies increasingly on computer generated images.

Schummer (2005) instead astutely combined network analysis and content analysis using Amazon.com data to identify the networks of books that the public reads about nanotechnology in the US. His findings suggest that the broader public connects nanotechnology to visions about dissolving the human/machine distinction.

However, there are no studies on how nanotechnology’s information is shared on the social web. The increasing importance of the Web for information seeking indicates the significance of studying this public arena and presents enormous potential in data collection as much as difficult methodological challenges.

4. Methodology

There is an ongoing debate on how standard content analysis should be adapted to web content (Herring, 2010) and this paper adopts a mixed research design. It uses web metrics data about the general pattern of information sharing on nanotechnology and an automatic text analysis on the content of tweets. In addition, it performs “sentiment analysis” to identify the general emotional charge – in terms of the negative or positive valence – of the corpus.

This study explores the use of social media to share and discuss information about nanotechnology and provide data on the following research questions:

Research question 1: What are the strategies to gather and share information by Twitter about nanotechnology?

Research question 2: What are the main drivers of these conversations?

Research question 3: How is nanotechnology anchored in this social medium?

Research question 4: What is the overall “sentiment” (where sentiment indicated the overall polarisation of tweets in terms of negative or positive emotions) of the tweets?

Sampling

The sampling procedure adopted the criteria of data collection based on their content feature and their time of publication. The study tracked randomly selected tweets for 60 days in 2011 with the procedure of collecting tweets containing the use of the keywords and hashtags “nanotechnology”, “nano” or “nanotechnologies”. The unit of analysis was a tweet and a major constraint on the sampling strategy was the computing power needed to process a very large number of tweets. The following strategy was adopted: in 2010 two tests were carried out, monitoring one week in September and one in December: both tests produced an average of 2600 tweets on nanotechnology per month. The upper computational limit of the two types of software used in the analysis roughly corresponded to 30,000 tweets. Hence, 60 days in 2011 were selected using a random calendar date generator.

A hashtag is a tag embedded in a message posted on Twitter, consisting of a word within the message prefixed with a hash sign (#). Although the use of these previously mentioned hashtags occasionally captured tweets that were not written in English, the corpus was restricted to English tweets in order to be analysed using the lemmatisation and automatic coding described in the next section. The source of data was the public API (application programming interface) of Twitter using a free for non-commercial use version of the proprietary service of tracking public tweets based on a Python script. The total number of tweets collected during the 60 days of tracking is 24634 (N = 24634). The tracking algorithm includes the following information: content of the post, username and number of followers.

Procedure

Data were first analysed using descriptive web metrics in terms of basic characteristics of the tweets and the user’s overall behaviour. The second step was to aggregate all tweets in a new and distinct corpus in order to analyse it using text-mining techniques. The corpus was prepared for quantitative text analysis using the software T-Lab 7.3 and IBM Text Analytics. Automatic lemmatisation was performed with a custom dictionary, thus the vocabulary of the corpus was checked to disambiguate homographs. In addition, the resulting keywords and those excluded were checked to avoid false positives. An important decision regarded the text segmentation strategy and therefore the selection of the unit of context used in the analysis. Three options were available: sentences, chunks or paragraphs. The unit “sentence” was considered the best given the short nature of tweets. More precisely, an elementary context as a sentence was every sequence of words interrupted by a full stop and carriage return, whose dimensions are up to 1000 characters (the limit for a tweet is 140).

The analysis proceeded in the following way. First, it constructed a data table composed of context units by lexical units with presence/absence values. Secondly, it applied TF-IDF (term frequency–inverse document frequency) normalisation and scaling of row vectors to unit length (Euclidean norm). Then it clustered context units using as measure cosine coefficients and as method bisecting K-means.² For each of the obtained partitions, it constructed a contingency table of lexical units by clusters and it applied a chi square test to all the intersections of the contingency table. The last step was to perform a correspondence analysis of the contingency table of lexical units by clusters. Because the algorithm used (bisecting K-means) produces a hierarchical clustering, there are several potential solutions in terms of cluster partitions. Thus, another important decision required setting a maximum number of clusters to be obtained. These are actually constraints in the sense that allowing for twenty clusters does not necessary mean that all of them will be produced because the algorithm will always minimise the number of clusters. After testing with two limits, ten and twenty, it produced the same results, indicating that the clusters obtained were stable. Subsequent to computing the quantitative semantic analysis, semantic clusters were interpreted by going back to text fragments of the articles referred by the most highly associated lemmas.

Sentiment analysis

The sentiment polarisation classification was operationalised according to the following procedure using IBM Text Analytics.

The first step was to create “concepts”. During the extraction process, the text data are scanned and analysed in order to identify interesting or relevant single words or phrases in the text. These words and phrases are collectively referred to as “terms”. Using the linguistic resources, the relevant terms are extracted and then similar terms are grouped together under a lead term called a “concept”.

The grouping is based on two techniques: concept root derivation and concept inclusion. The concept root derivation technique creates categories by taking a concept and finding other concepts that are related to it by analysing whether any of the concept components are morphologically related, or share roots. This technique is very useful for identifying synonymous compound word concepts, since the concepts in each category generated are synonyms or closely related in meaning.

The concept inclusion technique builds categories by taking a concept and, using lexical series algorithms, identifying concepts included in other concepts. The idea is that when words in a concept are a subset of another concept, it reflects an underlying semantic relationship. For example, if you have the concept “continental breakfast”, which has the component set {breakfast, continental}, and you have the concept “breakfast”, which has the component set {breakfast}, the algorithm would conclude that continental breakfast is a kind of breakfast and group these together.

When the concept root derivation or the concept inclusion techniques are applied, the terms are first broken down into components (words) and then the components are de-inflected. When a technique is applied, the concepts and their associated terms are loaded and split into components based on separators, such as spaces, hyphens, and apostrophes. For example, the term “system administrator” is split into components such as {administrator, system}. However, some parts of the original term may not be used and are referred to as stop words (e.g. and, as, by, of, etc.).

A second order of identification proceeds in creating “types” out of concepts.

Types are semantic groupings of concepts. When concepts are extracted, they are assigned a type to help group similar concepts. Several built-in types are delivered with IBM SPSS Text Analytics, such as <Location>, <Organization>, <Person>, <Positive>, <Negative> and so on. For example, the <Location> type groups geographical keywords and places. This type would be assigned to concepts such as Chicago, Paris, and Tokyo. Concepts that are not found in any type dictionary but are extracted from the text are automatically typed as <Unknown>.

The sentiment types are semantic groupings of concepts based on their emotional valence. Such emotional valence is associated using Sentiwordnet 3.0 (Esuli and Sebastiani, 2006), an enhanced lexical resource explicitly devised for supporting sentiment classification and opinion mining applications (Pang and Lee, 2008).

5. Findings

This section will first present a set of web metrics values followed by the latent semantic analysis and the sentiment analysis performed on the entire corpus of tweets (N = 24634).

Web metrics

The types of tweets collected in the corpus were 92% new messages, 7% “re-tweets” (a message forwarded to each of the user’s followers) and 1% “mentions” (or quotes, usually associated with replies). Perhaps unsurprisingly, 94% of tweets contained a link to a website. The high number of links to external websites indicates a trend for “forwarding” behaviour rather than producing only single comments. On Twitter this kind of behaviour is not unusual. Two concepts are important in evaluating tweets and users: reach and exposure. Reach is the total number of unique Twitter users who received tweets about the search term. Exposure is the total number of times tweets about the search term were delivered to Twitter users. These two metrics can be applied to users as well. The exposure of a user is his/her number of followers, the reach is the number of unique followers compared to those of other users.

Table 1 summarises the aforementioned metrics for the corpus on nanotechnology. The total number of posts was 24,634 made by 10,030 users, the total exposure was 6,685,493 while the reach was 3,335,870.

Table 1.

Breakdown of number of posts, reach and users.

Totals	Count
Total reach	3 335 870
Total posts	24 634
Total people	10 030

Number of posts	Number of users

1	6272
2 to 3	2234
4 to 5	681
> 5	843

Users by number of followers

Number of followers	Number of users
0 to 100	7649
100 to 500	1503
500 to 1000	373
1000 to 2500	282
> 2500	223

Using these values we can compute the reach/exposure ratio that represents the depth of penetration of tweets about a topic. A lower reach/exposure ratio suggests that people are seeing tweets about a topic over and over, while higher ratio numbers suggest broad but shallow penetration of that topic.

A moderate or normal ratio will be anywhere between 0.2 and 0.59 (Bakshy et al., 2011). This suggests a power law distribution of tweets, re-tweets and amplification. In this case, some people are tweeting multiple times; some influencers are tweeting to lots of followers; and most people are tweeting once or twice to their smaller set of followers. In the case of the nanotechnology corpus, the ratio was 0.498 indicating a moderate ratio and a Pareto cumulative distribution³ – this is a feature common to small-world networks and the very nature of the World Wide Web (Albert, Jeong and Barabasi, 1999).

Figure 1 presents the top 25 most active users showing a typical power law distribution in the number of tweets per user. The list comprises a mix of dedicated individual users and of Twitter channels dedicated to nanotechnology news. However, the most active users appear to be individuals rather than the official channels of scientific institutions or organisations. It may be noted that Figure 1 confirms what has been described before about the nature of the tweets’ distribution in terms of user activity: the pattern represents a Pareto distribution in which a few users are very active while a large number of users posted only one or two tweets.

Figure 1.

Top 25 users according to activity (number of posts about nanotechnology) and top 25 users according to number of followers (exposure).

Figure 1 presents the top 25 users that tweeted on nanotechnology in terms of their exposure (or number of followers). The top two tweeters in terms of number of followers and therefore those with the highest exposure are the Twitter channel of a TV channel dedicated to science – ScienceChannel – and an algorithmically constructed channel (tw_top_science) that selects and re-tweets science news from various official sources (such as Nature, Scientific American, NASA, etc.). The next two are science fiction fans with a large number of followers – “Duncan Jones” is also the director of two science fiction movies – and the fifth is the tweet channel of the well-known science fiction writer William Gibson. As in the case of the most active users, the number of followers is distributed according a Pareto distribution.

The next step is to analyse the content of the tweets using latent semantic analysis to identify the main semantic domains present in the overall corpus.

Latent semantic analysis

The algorithmic grouping of tweets in semantic domains according to the procedure described in the methodology section revealed a total of seven clusters (Figure 2).

Figure 2.

Clusters obtained from the detection algorithm of elementary contexts from the overall corpus (lemmas, N = 9199). Geometric proximity/distance stands for semantic proximity or distance.

The seven clusters were labelled as: “Fight Cancer”, “News/Conference”, “Mobile Phone”, “Science Projects”, “Applications”, “Opposition” and “Residual”.

The semantic cluster “Fight Cancer” accounted for 9.89% of the corpus. The cluster “News/Conference” was 18.6%, the cluster “Mobile Phone” 15.5%, “Science Projects” 15.5%, nanotechnology and a wide range of other applications 9.2%. The only semantic cluster containing opposition to nanotechnology (“Opposition”) was 15.2%. The residual cluster was 16% of the corpus and contained all the elementary contexts that could not be grouped in a statistically significant way by the algorithm of the software.

The cluster “Fight Cancer” – as suggested by its label – regards the application of nanotechnology to fight and cure cancer. The presence of a specific cluster indicates a significant emphasis on the medical application of nanotech and specifically for drug-delivery targeting tumour cells. Most of the tweets in this cluster were about news on potentially breakthrough medical treatments based on nanotechnology and this is the reason for the semantic proximity with the “News” cluster. Examples are “Breast Cancer: How Far Have We Come? nanotechnology 0 rads 1,000x more sensitive than mammogram http://t.co/0N2mcAs4 via @huffingtonpost”; “EGEN, Nanotechnology Characterization Laboratory Partner to Accelerate Human Clinical Trial of Brain Cancer Therapy http://t.co/dtF0CSJS”; “Cancer Nanotechnology: Methods and Protocols (Methods in Molecular Biology): Early detection of cancer at the c... http://t.co/X06uZEIK”.

The second cluster “News/Conference” contains tweets about public events related to nanotechnology; this is the biggest cluster, indicating that a large number of tweets are about promoting these conferences and public meetings. Examples of these sorts of tweets are: “REMINDER: Conference on applying #nanotechnology to save #environment in Taiwan http://t.co/NqJ6oVI1 #Taiwan”; “Hysitron will be exhibiting at the NanoTechnology for Defense Conference in Bellevue, Washington on 24-27 October. Visit us at Booth #212”; “The Nanotechnology Conferences http://t.co/FsJ2x0Sd #FutureConcepts”; “Trends in Nanotechnology (TNT2011) International Conference - Canary Islands (Spain): call for papers open - http://t.co/GcSy2dfB”; “Indians form body to boost nanotechnology http://t.co/q7d6l549”.

The third cluster is labelled “Mobile Phone” and it concerns nanotechnology’s application to mobile phones, in particular smartphones. As with the potential medical application of nanotechnology (see the previous cluster on cancer), its application on mobile phones was at the centre of a substantial amount of tweets standing out among the other gadget/goods oriented applications. Examples of these tweets are: “HzO Nanotechnology Seal Keeps Smartphone from Drowning 2012 CES Press Preview in NYC: http://t.co/DwEHL5XK”; “Nokia HumanForm Nanotechnology Smartphone Concept video Social Media News and viral video PkCrunch Mobile Phone ... http://t.co/xkBrU3Y4”; “Dial N for Nano -- how nanotechnology will improve smartphone battery life http://t.co/QYZ6qBcL #tech”.

The following cluster is “Science Projects” concerning current nanotechnology research and also a broad discussion of its potential. Examples of tweets in this group are: “Mapping Nanotechnology Innovations and Knowledge: Global and Longitudinal Patent and Literature Analysis (Integr... http://t.co/Ja2i0IJL”; “Nanotechnology is Driver to Take U.S. Out of Economic Doldrums, Says Nanofilm CEO, Scott Rickert”; “Sustainability for nanotechnology: NanoSustain: The NanoSustain project is investigating the life-cycle of sever... http://t.co/O6Uyu6Ga”; “Nanotechnology Now - Press Release: "EU project IMOLA starts ... http://t.co/7Dlsws1z”.

The cluster “Applications” refers to applications other than those for medicine or mobile phones. It contains a wide array of potential applications with a noticeable amount of agricultural applications. Here are some examples from the corpus: “Nanotechnology is Changing the Rules in Ceramics, says Nanofilm CEO Scott Rickert http://t.co/7qEqWBTj”; “Clear insulation coatings allow easy inspection of pipes. Possible w/ #nanotechnology http://t.co/E7g3z7AV”; “Nanotechnology Applications for Agriculture & Water Safety in Developing Countries. http://t.co/43LhEqt1”; “Agronomist Alberto Popper tells http://t.co/iRXvr2ZR about what makes nanotechnology tick and how it is implemented. http://t.co/BrzzdsxR”; “#Geek #News: Nanotechnology used to create a superhydrophobic surface for clothes. No stains and it’s just kinda cool. http://t.co/pwAGYlGH”.

The last cluster is about the risk and opposition to nanotechnology. Essentially there are three themes present in the cluster: the potential toxicity of nanotechnology for food and consumer goods, its military application and the “Pandora’s box” scenario of a technology that might run out of control. These are common themes in the risk discourse of nanotechnology found in many previous studies. Examples of tweets in this cluster are: “Dark side of nanotechnology http://twitter.com/NanotecYPM/status/136178184235393025”; “Why mankind must face and destroy the growing robot menace? Humans scared of robots? Nanotechnology is far more scary. http://t.co/5DHVFu56”; “Good things come in small packages? A chat about nanotechnology and food safety: Photo: Titanium dioxide nanopar... http://t.co/yOVFLMhd”; “Consumer Fact, Not Science Fiction: The Need To Keep Nanotechnology Products Safe http://twitter.com/NanotecYPM/status/136154858737778688”; “Risk vs. Reward: Why it’s important to thoroughly assess the toxic risks of nanotechnology http://t.co/MdDvXKWr #science #nanotechnology”; “Video: Nanotechnology Ushers in Cloak of Invisibility | Truth Is Scary http://t.co/IRGvYswV”.

The next step is to outline the findings from the sentiment analysis to see the overall emotional valence of the tweets contained in the corpus.

Sentiment analysis

As described in the previous methodology section, clusters were created of terms by the emotional valence of the words, such as adjectives and nouns that have either an average positive or negative value, of tweets and their co-occurrences. Figure 3 shows the semantic network of categories associated with an overall positive sentiment.

Figure 3.

Semantic network of nanotechnology in terms of contexts (semantic domains) and sentiments (negative and positive). Asterisks indicate the strength of sentiment polarisation, three asterisks being the strongest. The figure expands the details of the semantic network of “excellent” in terms of contexts (semantic domains) and sentiments.

As shown by the network, positive feelings about nanotechnology are related in particular to business, medicine and science in general. As discussed in the latent semantic analysis, the application for medical treatments (in particular a remedy for cancer) and for commercial goods (smartphones) showed a major salience in the corpus. Positive markers such as “excellent” and “well” suggest positive feelings.

In Figure 3 we explore in more detail the semantic network linked to “excellent”. From the rather dense network, it is possible to discern several contexts associated with the positive type “excellent” such as “medicine”, “apparel/clothes”, “chemistry”, “computer/computing”, “coating”, “physics/carbon nanotubes”, “home improvement” etc. The semantic network of positive terms related to nanotechnology confirms and substantiates the previous interpretation of the latent semantic analysis: a wide range of nanotechnology applications are represented as positive. It can be noticed that the strongest link is between the “excellent” and “almost improving” categories, indicating that nanotechnology is firmly within the inevitable scientific and technological progress narrative.

Next, we examine the semantic network associated with negative sentiments represented in Figure 3. The context is that negative tweets about nanotechnology were few in number. Most negative sentiments were expressed in the form of: “unknown problem” (meaning unknown foreseeable problem); “not safe”; “needs more” understanding; “mad/crazy” referring to the dystopian and self-destructive scenario into which humankind is approaching; “hype”, a lower tone of critique regarding the unjustified hype around nanotechnology and its promises (or claimed promises). There is also a strong link with “would be good”, which relates to the scepticism surrounding the virtually risk-free applications of nanotechnology.

Overall, the semantic network of negative sentiments reflects the content of the semantic cluster identified with the latent semantic analysis and labelled as “Opposition” in the previous analysis. The negative feelings in tweets about nanotechnology appear to be mainly associated with the unknown risks of nanotechnology, in particular the toxicity of nanoparticles, and side effects of its application to consumer goods rather than the most catastrophic scenarios that are nevertheless present in the corpus.

6. Discussion

The findings from the web metric analysis of the tweets on nanotechnology indicate that, overall, some people are tweeting multiple times, some influencers are tweeting to many followers, and most people are tweeting once or twice to their smaller set of followers. The most active users appear to be individuals rather than the official channels/representatives of scientific institutions or organisations, however, the latter are, in general (and perhaps unsurprisingly) the most followed users. In both cases, the most active and followed users are described by a Pareto distribution. The low percentage of “re-tweets” and “mentions” indicates that nanotechnology is not an object of conversation on Twitter but rather that Twitter is another channel of diffusion for nanotechnology (Boyd, Golder and Lotan, 2010).

Moving to the content of the tweets, there is a remarkable similarity with the findings of previous studies on the discussion and representation of nanotechnology in the national press of different countries (e.g. Anderson et al., 2005; Kjølberg, 2009; Kjærgaard, 2010; Donk et al., 2012; Groboljsek and Mali, 2012; Veltri, 2012). Perhaps it is not surprising, given the nature of the medium, that the large majority of tweets were dedicated to reporting news of nanotechnology events or nanotechnology applications for medical research and consumer goods. Little emphasis was placed on resources to understand the science and technology involved. The tweets about concerns and opposition to nanotechnology were limited (around 15% of the corpus) and mainly focussed on the issue of potential toxicity of nano-materials and the military applications of nanotechnology (for example “the cloak of invisibility” for soldiers and spies). Related to content is the sentiment analysis which indicates predominantly positively loaded words in the corpus. Negative sentiments mainly took the form of uncertainty and fear of the unknown rather than open hostility.

The generalisability of this study is limited by three main restrictions that impacted upon its research design: limiting the sample of tweets to the English language only, the relatively short period of monitoring and tracking (60 days) – although this is fairly standard with these sorts of studies, and the characteristics of Twitter users in comparison to those of the general public or the average Internet user. The first two limitations are caused by the lack of necessary computing power to process and analyse an extremely large number of tweets. For example, tracking over one year might have resulted in about 300,000 tweets in English alone, there is no way to tell how large this number might be if other languages were also included.

Data from the numerous waves of data collection conducted by the PEW Research Center (2010) indicate some of the groups who are notable for their relatively high levels of Twitter use:

Internet users aged 18–29 are significantly more likely to use Twitter than are older adults.

Urban residents are roughly twice as likely to use Twitter as rural dwellers.

Women and the college-educated are also slightly more likely than average to use Twitter.

On a more general note, individuals who use ICT (information and communication technology) tend to have larger and more diverse social networks, providing evidence against fears of their being socially isolated individuals. On average, the size of people’s discussion network – those with whom they discuss important matters – is larger among ICT users compared to non-users (Hampton et al., 2009).

Java, Song, Finin and Tseng (2007) investigated motivations behind tweeting using computational linguistics techniques and confirmed that Twitter is mainly used to communicate daily activities and to seek or share information about a given topic.

7. Conclusions

This study combines a mature technique of semantic clustering with sentiment and web metrics analyses. The choice and combination of such methodologies mirror the research questions defining the boundaries of this study. The overall approach is original because it focuses on the “representational space”, the implicit frames and the social representations of nanotechnology diffused through a social medium rather than investigating only the medium, its users and their structural properties, which is usually the case in many recent studies about the social web.

The semantic clustering represents the main analysis with the web metrics and sentiment analyses as additional and contextual data. This strategy of analysis aims at answering the main research question: How is nanotechnology anchored in this social medium? Additional analysis informed by web metrics on the use of Twitter as a medium represents useful information to understand nanotechnology news diffusion in the social web. Web metrics provide insights about users and the structure of Twitter’s “nanotechnology network” but an exhaustive study of the latter goes beyond the scope of this analysis. For example, a social network analysis could determine the network roles and characteristics of users (e.g. centrality) in the sharing and discussing of a topic. In this context, the main research question would be: Who are those that spread and collect nanotechnology’s information on Twitter (or whichever social medium) and what are their networks’ characteristics? However, a social network analysis of this kind poses several methodological challenges, for example it requires a longitudinal dimension of data collection and related computing power that are not easy to access.

Beside web metrics, sentiment analysis provides additional information on emotional polarisation associated to the semantic clusters and therefore present in the sharing and discussing of nanotechnology on Twitter. This information adds a further evaluative dimension of the semantic space, something that is important because the social web is above all a medium of opinion exchange. As much as for the network structure, emotional polarisation could easily be a study in itself with a more refined sampling strategy and in-depth analysis. This sort of study is a promising perspective in the studying of public opinion, given that there are few doubts on the role of emotions in guiding the assessment of technologies and their risks and benefits (Roeser, 2010).

The findings suggest that nanotechnology is not an object of conversation on Twitter but rather that information is shared and diffused by a limited number of “power users”, highly “followed” and “linked”. The content of the tweets indicates a semantic space that mirrors largely the one of traditional media outlets such newspapers. However, the combination of semantic and sentiment analysis reveals that the negative emotions are more related to uncertainty rather than hostility.

The methodology and analysis applied in this study could monitor the status of the conversation of nanotechnology (or other technologies) on a given social medium in order to detect the origin, growth and diffusion of semantic constructs and relative emotional polarisations. Adding a geospatial tagging could also help the precision and interpretation of such monitoring and even lead to predictions combined with other social variables (see the large-scale example of Leetaru, 2011).

This study represents a first attempt to use the social web as a source of data in public understanding of science research, offering a discussion about its potential and on the necessity of establishing a solid set of methodologies.

There can be few doubts, we believe, as to the growing importance of the social web in the study of the public understanding of science. These new arenas constitute a vast pool of potential data for analysing public opinion dynamics regarding the public understanding of science and risk perception. Social media are now a key source of data as new opinion-tracking methods such as mining, aggregating and analysing online data in real time have become available.

In the near future, the first objective will be to develop a standard methodology capable of dealing with the complexity and scope inherent to this kind of analysis, while the long-term goals are to analyse specific topics using online data and to examine real-time reactions to specific events. As mentioned above, it is possible to apply recent techniques of online opinion mining and social network analysis of the diffusion routes and evolution of opinions and ideas (memes). This approach could lead to the development of a memetics of public understanding of science and risk perception (Weng et al., 2012). There are exciting opportunities ahead that need thorough theoretical and methodological debate, as is always the case when new opportunities arise.

Footnotes

Funding

This research received no specific grant from any funding agency in the public, commercial or not-for-profit sectors.

Notes

Author biography

Giuseppe Alessandro Veltri holds an MSc in Social Research Methods from the Methodology Institute of the London School of Economics (LSE) and a PhD in Social Psychology from the LSE. He is currently Lecturer in Media and Culture at the School of Political, Social and International Studies of the University of East Anglia. His research interests focus on public opinion research, public understanding of science, social representations and social psychology of economic life.

References

Albert

Jeong

Barabasi

(1999) Internet: Diameter of the World-Wide Web. Nature 401: 130–131.

Anderson

Allan

Petersen

Wilkinson

(2005) The framing of nanotechnology in the British newspaper press. Science Communication 27: 200–220.

Bader

(1990) How science news sections influence newspaper science coverage: A case study. Journalism Quarterly 67(1): 88–96.

Bakshy

Hofman

Mason

Watts

(2011) Everyone’s an influencer: Quantifying influence on Twitter. WSDM ’11: Proceedings of the Fourth ACM International Conference on Web Search and Data Mining, Hong Kong, 9–12 February, ACM, pp. 65–74.

Boyd

Golder

Lotan

(2010) Tweet, tweet, retweet: Conversational aspects of retweeting on Twitter. HICSS-43, IEEE, Kauai, HI, 6 January.

Bucchi

(2002) Science in Society: An Introduction to Social Studies of Science. London: Routledge.

Byrne

Namuth

Harrington

Ward

Lee

Hain

(2002) Increasing public understanding of transgenic crops through the World Wide Web. Public Understanding of Science 11(3): 293–304.

Chew

Eysenbach

(2010) Pandemics in the age of Twitter: Content analysis of tweets during the 2009 H1N1 outbreak. PLoS ONE 5(11): e14118.

Delborne

Anderson

Kleinman

Colin

Powell

(2011) Virtual deliberation? Prospects and challenges for integrating the Internet in consensus conferences. Public Understanding of Science 20(3): 367–384.

10.

Donk

Metag

Kohring

Marcinkowski

(2012) Framing emerging technologies: Risk perceptions of nanotechnology in the German press. Science Communication 34(1): 5–29.

11.

Dunwoody

Scott

(1982) Scientists as mass media resources. Journalism Quarterly 59: 52–59.

12.

Esuli

Sebastiani

(2006) SENTIWORDNET: A publicly available lexical resource for opinion mining. Proceedings of the 5th Conference on Language Resources and Evaluation (LREC’06), Genova, Italy, pp. 417–422.

13.

Eveland

Dunwoody

(1998) Users and navigation patterns of a science World Wide Web site for the public. Public Understanding of Science 7(4): 285–311.

14.

Fahmy

Wanta

Johnson

Zhang

(2011) The path to war: Exploring a second-level agenda-building analysis examining the relationship among the media, the public and the president. International Communication Gazette 73: 322–342.

15.

Friedman

(1989) TMI: The media story that will not die. In: Walters

Wilkins

Walters

(eds) Bad Tidings: Communication and Catastrophe. Hillsdale, NJ: Lawrence Erlbaum, pp. 63–83.

16.

Friedman

Egolf

(2005) Nanotechnology: Risks and the media. IEEE Technology and Society Magazine 24: 5–11.

17.

Gamson

Modigliani

(1989) Media discourse and public opinion on nuclear power: A constructionist approach. American Journal of Sociology 95: 1–37.

18.

Gaskell

Bauer

Durant

(1998) The representations of biotechnology: Policy, media and public perceptions. In: Durant

Bauer

Gaskell

(eds) Biotechnology in the Public Sphere: A European Source Book. London: Science Museum Press, pp. 3–15.

19.

Gaskell

Ten Eyck

Jackson

Veltri

(2005) Imagining nanotechnology: Cultural support for technological innovation in Europe and the United States. Public Understanding of Science 14(1): 81–90.

20.

Gerbner

Gross

(1976) Living with television: The violence profile. Journal of Communication 26(2): 172–199.

21.

Gopfert

(1996) Scheduled science: TV coverage of science, technology, medicine and social science and programming policies in Britain and Germany. Public Understanding of Science 5(4): 361–374.

22.

Groboljsek

Mali

(2012) Daily newspapers’ views on nanotechnology in Slovenia. Science Communication 34(1): 30–56.

23.

Hampton

Goulet

Ja Her

Rainie

(2009) Social Isolation and New Technology. Pew Internet & American life Project. Available at: http://pewresearch.org/pubs/1398/internet-mobile-phones-impact-american-social-networks

24.

Hanson

(2012) Amidst nanotechnology’s molecular landscapes: The changing trope of subvisible worlds. Science Communication 34(1): 57–83.

25.

Hawn

(2009) Take two aspirin and tweet me in the morning: How Twitter, Facebook, and other social media are reshaping health care. Health Affairs 28(2): 361–368.

26.

Herring

(2010) Web content analysis: Expanding the paradigm. In: Hunsinger

Allen

Klastrup

(eds) The International Handbook of Internet Research. Berlin: Springer Verlag, pp. 233–249.

27.

Hoijer

(2010) Emotional anchoring and objectification in the media reporting on climate change. Public Understanding of Science 19(6): 717–731.

28.

Holliman

(2004) Media coverage of cloning: A study of media content, production and reception. Public Understanding of Science 13(2): 107–130.

29.

Jain

(2009) Practicing medicine in the age of Facebook. New England Journal of Medicine 361(7): 649–651.

30.

Java

Song

Finin

Tseng

(2007) Why we twitter: Understanding microblogging usage and communities. Proceedings of the 9th WebKDD and 1st SNA-KDD 2007 Workshop on Web Mining and Social Network Analysis, San Jose, CA, 12–15 August, ACM, pp. 56–65.

31.

Jones

(1997) The boffin: A stereotype of scientists in post-war British films (1945–1970). Public Understanding of Science 6(1): 31–48.

32.

Kjærgaard

(2010) Making a small country count: Nanotechnology in Danish newspapers from 1996 to 2006. Public Understanding of Science 19: 80–97.

33.

Kjølberg

(2009) Representations of nanotechnology in the Norwegian newspapers: Implication for public participations. Nanoethics 3: 61–72.

34.

Landau

Groscurth

Wright

Condit

(2008) Visualizing nanotechnology: The impact of visual images on lay American audience associations with nanotechnology. Public Understanding of Science 18(3): 325–337.

35.

Lang

(1983) The Battle for Public Opinion: The President, the Press, and the Polls during Watergate. New York: Columbia University Press.

36.

Leetaru

(2011) Culturomics 2.0: Forecasting large-scale human behavior using global news media tone in time and space. First Monday 16(9) 5 September. Available at: http://firstmonday.org/htbin/cgiwrap/bin/ojs/index.php/fm/article/viewArticle/3663/3040

37.

Leon

(2008) Science related information in European television: A study of prime-time news. Public Understanding of Science 17(4): 443–460.

38.

Lewenstein

Gorss

Radin

(2005) The salience of small: Nanotechnology coverage in the American press, 1986-2004. Paper presented at the International Communication Association, New York.

39.

Lewis

Kaufman

Gonzalez

Wimmer

Christakis

(2008) Tastes, ties, and time: A new social network dataset using Facebook.com. Social Networks 30(4): 330–342.

40.

Marwick

Boyd

(2011) I tweet honestly, I tweet passionately: Twitter users, context collapse, and the imagined audience. New Media & Society 13(1): 114–133.

41.

McCombs

Ghanem

(2001) The convergence of agenda-setting and framing. In: Reese

Gandy

Grant

(eds) Framing Public Life: Perspectives on Media and our Understanding of the Social World. Mahwah, NJ: Lawrence Erlbaum, pp. 67–81.

42.

McCombs

Shaw

(1993) The evolution of agenda-setting research: Twenty-five years in the marketplace of ideas. Journal of Communication 43(2): 58–67.

43.

Moscovici

(2000) Social Representations: Explorations in Social Psychology. Cambridge: Polity Press.

44.

Neidhardt

(1993) The public as a communication system. Public Understanding of Science 2: 339–350.

45.

Nelkin

(1994) Promotional metaphors and their popular appeal. Public Understanding of Science 3(1): 25–31.

46.

Ogden

(2001) The language of agricultural biotechnology: Terminate or be terminated. Organization & Environment 14: 336–340.

47.

Pang

Lee

(2008) Opinion mining and sentiment analysis. Foundations and Trends in Information Retrieval 2(1/2): 1–135.

48.

Pew Research Center (2010) Global Publics Embrace Social Networking. Available at: http://pewresearch.org/pubs/1830/social-networking-computer-cell-phone-usage-around-the-world

49.

Roeser

(2010) Emotions and Risky Technologies. Delft: Springer.

50.

Rogers

Marres

(2000) Landscaping climate change: A mapping technique for understanding science and technology debates on the World Wide Web. Public Understanding of Science 9(2): 141–163.

51.

Scheufele

Lewenstein

(2005) The public and nanotechnology: How citizens make sense of emerging technologies. Journal of Nanoparticle Research 7: 659–667.

52.

Schummer

(2005) Reading nano: The public interest in nanotechnology as reflected in purchase patterns of books. Public Understanding of Science 14(2): 163–183.

53.

Shirky

(2010) Cognitive Surplus: Creativity and Generosity in a Connected Age. New York: Penguin Press.

54.

Triunfol

(2004) Dynamics of list-server discussion on genetically modified foods. Public Understanding of Science 13(2): 155–175.

55.

Veltri

(2012) Viva la nano-revolución! A semantic analysis of the Spanish national press. Science Communication, doi: 10.1177/1075547012440353

56.

Veltri

Suerdem

(2011) Worldviews and discursive construction of GMO-related risk perceptions in Turkey. Public Understanding of Science, doi: 10.1177/0963662511423334

57.

Waldrop

(2008) Science 2.0: Great new tool or great risk? Scientific American May. Available at: http://www.scientificamerican.com/article.cfm?id=science-2-point-0

58.

Waters

Burnett

Lamm

Lucas

(2009) Engaging stakeholders through social networking: How non-profit organizations are using Facebook. Public Relations Review 35(2): 102–106.

59.

Weng

Flammini

Vespignani

Menczer

(2012) Competition among memes in a world with limited attention. Scientific Reports 2: Article number 335.

60.

Westling

(2007) Expanding the public sphere: The impact of Facebook on political communication. Society 28: 835–860.

61.

Wilkins

Patterson

(eds) (1991) Risky Business: Communicating Issues of Science and Risk, and Public Policy. New York: Greenwood.

62.

Zappavigna

(2011) Ambient affiliation: A linguistic perspective on Twitter. New Media & Society 13(5): 788–806.