Opinion Leaders in the Digital Age – Social Network Analysis for Renewable Energies on Twitter

Evaluating opinions, sentiments and relationships within groups become increasingly easy due to wide availability of data on social media platforms. This article discusses aforementioned aspects for a Twitter network consisting of 37,760 German tweets related to renewable energies which were collected for the year 2019. Besides identifying the most frequent words like energy revolution, CO₂, power, Germany and coal, this paper also evaluates the reach and importance of specific tweets and individuals. While the follower count of any given user struggles to represent their true influence, the farthest-reaching tweets do indicate opinion leaders which is supported by the LeaderRank-Algorithm.

1  Introduction

Many politicians utilize social media to further their agenda, the most prominent example being Donald J. Trump who appointed Twitter as his favorite tool to share his own, unfiltered opinion. However, the shift from real world discussions to an online environment does not only affect the direct discourse between politicians. In fact, companies, NGOs and even individuals recognize the vast potential of social media platforms as they allow quick dissemination of any given opinion and thus provide a fairly cheap way of receiving attention.

With 340 million active users per month, Twitter is the 13th biggest social media platform worldwide (1) while being the 7th biggest in Germany behind WhatsApp, YouTube, Facebook, Instagram, Facebook Messenger and Pinterest (2). While it may be subject to discussions whether simplifying a social media platform to its active monthly users truly captures the relevance of any given platform, it still outlines its reach. In Germany, Twitter has 2.8 million active users per month of which 1.4 million use it on a weekly and 19 % on a daily basis (3). The differentiating aspect between Twitter and its counterparts with more monthly users is the fact that many important opinion leaders like politicians and activists prefer Twitter for sharing their thoughts. Twitter has a key influence on public opinion because “powerful multipliers […] use the platform and messages often end up in high-reach mass media (3). ” Thus Twitter has the status of a platform for gatekeepers. With this prevalence of thought leaders, the question who exerts the most influence becomes pressing.

The concept of opinion leadership builds upon the ideas of Katz and Lazarsfeld who dealt with the propagation of information through mass media (4). They pointed out that major parts of society are not influenced by mass media itself but rather by a few highly influential individuals. Generally, there are three characteristics that promote an individual into the position of an opinion leader. These are 1. Personification of Values, which refers to the similarities between an opinion leader and the individuals that they influence, 2. Competence or what an individual knows and whether they are perceived as a trustworthy authority due to their work expertise and 3. Strategic Social Position which refers to the position in a social system and how well they are connected (4). While opinion leaders are competent and influential (5), their competence is not necessarily exerted or pointed out actively, instead people unconsciously become more receptive to the ideas of opinion leaders (6). Opinion leaders direct and initiate change instead of acting themselves (7). Opinion leadership is not universally applicable, being a thought leader in one domain does not inevitably result in being a thought leader in all domains (8). Furthermore, there is a degree to which an individual, with relative frequency, influences opinions and behaviors of other individuals (9).

2  Research problem and objective

Opinion leaders inspire action in various fields, they can influence buying and consumption decisions but also change the direction of the public opinion in political matters. This is actively used in corporate communications, especially in marketing activities when opinion leaders or “influencers” are involved to directly advertise products to a target audience. In this context, it is not only often uncertain how the opinion leaders can be integrated in order to achieve the strategic goals of the communication activities. Frequently opinion leaders are unknown and not identified in the first place. Being unaware of thought leaders in one’s own domain inevitably results in an inability to recruit and cooperate with them. Even if the goal is not to cooperate, identifying opinion leaders is still crucial when it comes to research and uncovering additional information about the topic, monitoring the public discussion and identifying relations between the influencers. The lack of knowledge can lead to competitive economic disadvantages or it can limit how far one can spread ideas.

While developments towards renewable energies and a more sustainable future are prevalent in German politics, there still prevails nescience when it comes to the individuals that define and shape the public discourse. Besides the proponents of a drastic shift in the energy sector there also exist opponents that actively express their resentment towards changes. This article, intends to shed light on the sentiment, extend and reach of discussions about renewable energies in Germany by analyzing the discussion on Twitter using a network analysis. In turn, we analyze tweets from 2019 with a special focus on how geopolitical events may be reflected in an increased Twitter activity, we evaluate word frequencies, sentiment and try to identify opinion leaders and their connection to each other.

3  Methodology

A social network analysis captures relations within a given social network. The networks are defined by their nodes, in this case the Twitter users, and the edges between them which come into existence when one user mentions another. This ability to create edges towards other nodes is a prerequisite in this domain (10). A network is thus defined by a distinct amount of nodes and the sum of edges between them (11). Additionally, the collection of nodes and edges as a whole can be used to interpret the social behavior of participating individuals. (12)

Data acquisition was conducted using the python tool Twitterscraper (13) since access to historical Twitter data through the Twitter API has, in its free version, strict limitations. API is short for Application Programming Interface, it allows communication and exchange between two different systems. Even though Twitterscraper is available free of charge and not officially supported by Twitter it nevertheless proved valuable in previous research (14) and is thus judged to be a reliable way of acquiring the necessary data.

4  Data acquistion

The dataset for 2019 was collected on the 21st January 2020. Initially 8,195 Tweets containing the phrase “erneuerbare Energie(n)” (reneweable energies) were discovered. To further extend this dataset the words “erneuerbar” and “erneuerbare” were also searched on Twitter which increased the total dataset by 3,083 and 34,507 tweets respectively, resulting in a total dataset size of 45,785 German tweets from 2019. This size was deemed sufficient as it allows to make relevant conclusions about the topic without falling into a tiny niche which may only represent heavily biased parts of the discussion on renewable energies. Therefore, the data-set was not extended any further by including other hashtags or additional phrases.

While Twitterscraper allows to uncover historical data, it does not provide the full range of metadata. As a result, the unique tweet IDs were extracted from the original dataset and then send as individual requests to Twitter’s API. By using this two-step approach, it is possible to initially acquire historical data and then enhance it with the necessary metadata. As a result, it is possible to surpass Twitter’s API limitations with regard to the historical data and still receive the benefits of it. However, not all API requests were successful even after repeatedly trying. As there is no insight into the actual Twitter API this is a loss that at this point cannot be avoided. In combination with occasional duplicates in the dataset there remains a total of 37.760 tweets and their metadata which can be used for the final analysis.

5  Results

5.1  General data set metrics

In the network 13,287 unique accounts have been active and on average each account published 2.84 tweets. Cumulatively, all acounts have 52,182,970 followers, yet the actual number may be much smaller due to the fact that users usually follow multiple accounts and thus one person may follow 100 of the accounts that have been active in the network.

Out of the 37,760 tweets 17,715 contained a URL of which 14,228 were unique and out of those 167 were YouTube links which can be seen as a measure for the occurrence of videos within the network. In total 6,294 images were contained in the dataset. All the tweets accumulated 232,114 likes and 70,488 retweets which again can include likes and retweets from the same accounts, thus these metrics do not allow reliable conclusions about the total reach and impact of the network.

5.2  Tweet timeline

On average 712 tweets linked to “renewable energies” have been posted per week. This activity peaked on six different occasions leading to over 1,000 tweets a week as can be seen in figure 1. The origin of these peaks can partially be mapped to real world news events. In week 38 German political party CDU presented its climate agenda (15) which received harsh criticism due to the plans for the transportation sector (16). Simultaneously, Greta Thunberg and Fridays for Future recieved the “Ambassador of Conscience“ award by Amnesty International (17), Thunberg visited the United States (18), the Fridays for Future movement drove attention towards climate change with protests across the globe (19) and the German coalition agreed on a climate protection package (20).

5.3  Word frequencies

The text data of all 37,760 tweets needed to be prepared before a proper analysis was possible. The first step was to remove numbers and stop words from the text corpus. Stop words are words that occur frequently but contain little to no topic relevant content, they can be found in most texts and thus provide no actual information about texts. Stop words were filtered using the “stopwords” R package (21) and the stop word dictionary by Graham (22). Remaining but irrelevant words like the frequent occurrence of shortened Twitter URLs (t.co), ULR strings like https and numbers from 1 to 100 have been manually removed.

The finalized word frequencies can be seen in figures 2 and 3. The seemingly most important word “Energiewende”, which references the change from non-renewable energies to renewable sources, occurred over 4,000 times followed by “CO₂” and “Strom” (power) which occurred around 3,000 times. Overall, word frequencies of the Twitter debate indicate a strong emphasis on the energy revolution in relation to electricity (Figure 2).

Heat generation seems to be of little importance. Just the words coal and gas occur within the 20 most used words while they can both also be used to create electricity (Figures 2, 3).

Within the domain of renewable energies, wind power seems to be a dominant topic of discussion as there occur three words that are linked to it. Solar energy seems to play a secondary role in discussions. The CDU is the only political party that occurs frequently in our data corpus which may be linked to their presentation of a new climate agenda as well as their overall size and significance in German politics in general.

5.4  Potential opinion leaders by followership and most influential tweets by likes and retweets

To identify the most influential individuals two approaches have been pursued. Firstly, their potential reach and influence was judged by their total follower count. Secondly, tweets were identified that reached the highest cumulative count of retweets and likes, since influence drives interaction and retweeting and liking are the most common forms of interaction on Twitter. Liking a tweet and retweeting are different actions that could receive varying values but distinguishing between the two requires unique weights for both. Defining these weights and the reasoning behind it is a complex endeavor which cannot be detailed right here, thus likes and retweets receive equal weights and are summed up.

The 50 accounts with the highest follower count were sorted into seven distinct categories: news, politics, public figures, journalists, environmental organizations, businesses and others (Figure 4).

It becomes evident that news outlets and politics clearly dominate and have the most followers. However, this is not due to the fact, that all of them are thought leaders for renewable energies. It is rather due to the fact, that they pursue a wide range of interests. News organizations cover varying topics, the same applies to political parties and individual politicians. Furthermore, it is difficult to identify why any given individual follows a specific account, it may be due to an interest in renewable energies but it could also be an interest in any other domain that the respective Twitter account covers. Therefore, the follower count does not reliably display whether an account can be deemed as thought leader in the domain of renewable energies.

The top list of news accounts is being led by “Tages-schau” (-@-tagesschau), “Der Spiegel” (@-derspiegel), “Zeit-Online” (@-zeitonline), “Bild” (@-bild) and “Süddeutsche Zeitung” (@-sz). No companies linked to renewable energies were present, the only company that tweeted about this topic and made it onto the list was KLM, an airline from the Netherlands. In the political domain parties like “SPD” (@-spd), “CDU” (@-cdu) and “Die Linke” (@-dielinke) posted tweets in our network as well as the individual politicians Christian Lindner (@-c_lindner), Sebastian Kurz (@-sebastiankurz), Hans-Christian Ströbele (@-mdbstroebele), Sigmar Gabriel (@-sigmargabriel) and Peter Altmaier (@-peteraltmaier). Lastly, the city of Munich (@-stadtmuenchen) as well as the Ministry of Foreign Affairs (@-auswaertigesamt) are also present with a lot of followers.

At least in our case, the follower count itself does not indicate the actual reach of any given Twitter user, yet likes and retweets are measurable interactions that a tweet generated and thus indicate how influential that tweet was. Furthermore, both likes and retweets extend the reach of a tweet as they are being shown on the timeline of the user who retweeted it. In that way, likes and retweets should represent the actual influence far better than just the follower count. When ranking the tweets according to their total like and retweet count (Table 1) it becomes evident that none of those tweets was posted by an account from the previous top 50 follower count list.

Of the 30 most relevant tweets judged by like and retweet count, seven were posted by Volker Quaschning (@-vquaschning) a professor for renewable energy systems at the university of applied sciences for technology and economics in Berlin. Five tweets were posted by Fridays for Future Germany (@-fridayforfuture) and two were published by Sebastian Hornschild (@-fshhornschild), a city council in Erlangen who focuses on climate politics and is member of Fridays for Future. Luisa Neubauer (@-luisamneubauer), also a member of Fridays for Future, posted the tweet with the most likes (6.594) and retweets (2.222) (Table 1, Figure 5).

Most of the tweets seen in figure 5 were posted by accounts that are seen as authorities by the renewable energies Twitter community and that are mainly active in the domain of renewable energies. Therefore, it is no surprise that none of the accounts with the most followers made it to the list which might be due to their rather widespread profile. Retweets are not only a strong social component (29) they are also used to take a position in relation to the given network (30). There are three main aspects influencing the probability of a tweet being retweeted. Firstly, the individual aspects of the user confronted with the tweet (31, 32). Secondly, the source of information (33, 34) and lastly the tweet itself (35). However, the amount of valid information contained in the tweet is not necessarily the determining aspect since the expression of an opinion may have a stronger impact (36). In the analyzed network it is more likely to find tweets posted by active members that regularly participate in the online discussion compared to news outlets which can rather be seen as an external source that covers the subject of renewable energies without taking a stance themselves. Lastly, the most relevant tweets may also be linked to the prevalent media attention that Luisa Neubauer, Sebastian Hornschild as well as the Fridays for Future movement at large received in 2019.

5.5  Potential opinion leaders: graphs and LeaderRank

To visualize the entire renewable energies network of 2019, each individual user is defined as a node and mentioning other users result in edges to the respective nodes. As seen in figure 6, it becomes evident that many users are talking about renewable energies but are actually not integrated or linked to any other user in the network. This could be either due to the fact, that they actually just post tweets without participating in active discussions or it may result from the fact that not all discussion tweets contain the exact phrases that we searched for. Thus, an individual may be connected to another, but the relationship is not captured since the dataset does not include the tweet linking both together. In turn, these non-participating individuals found mainly in the periphery were filtered out since an individual not facilitating any interaction (replying to someone or being replied to be someone within the network) with other accounts possess little value when it comes to the analysis of a social network where the social context is of utmost importance.

Subsequently, as the graph is still quite crowded, low activity accounts with an indegree less than or equal to five were removed. Indegree captures how many ingoing edges a node has or, in our case, how frequently an account was mentioned by other nodes within the network. While this node reduction makes the graph less crowded and thus a little clearer it can also create nodes that are again on their own without connections to other nodes. This happens when an account is mentioned more than five times but the mentioning accounts were filtered out. Figure 7 displays a filtered sector of the original graph.

The size of the nodes growths with an increase of incoming edges (indegree) and the colour depends on their outdegree, darker colours represent a high outdegree (an account mentioning many other accounts) whereas lighter colours indicate low outdegree. Even in this filtered version it seems crowded and unstructured which is a result of the dataset size. However, this version or rather non-static interactive version of the graph can be used to not only answer subsequent questions but also to identify key players and how they interact as well as who interacts with them. Furthermore, a visual representation can be compared to graphs of future datasets to illustrate and analyze changes throughout different periods.

In the visualization Peter Altmaier (@-peteraltmaier), Volker Quaschning (@-vquaschning), Fridays For Future (@-fridayforfuture) as well as Luisa Neubauer (@-luisamneubauer) seem to be central nodes which indicates that they are opinion leaders in the field of renewable energies. Whether these users can be reliably identified as opinion leaders is determined with the LeaderRank algorithm. LeaderRank is an adaption of Googles PageRank algorithm (37), an iterative algorithm developed by Lü et. al. (38). The core idea of PageRank is to identify authorities and hubs. An authority in a network is frequently mentioned by other nodes whereas hubs are nodes that have a rather high outdegree and mention many other nodes, but this analysis clearly focuses on identifying authorities (opinion leaders). Additionally, opinion leadership is not just referenced by the indegree. Instead, it emphasizes that an opinion leader will also be mentioned by others. If a seemingly unknown node gets frequently mentioned by a highly influential node then this unknown node will be rewarded for the incoming edge from the opinion leader. The major difference between PageRank and LeaderRank (Equation 1) is the root node s_g which ensures the convergence of the final LeaderRank scores. In the first iteration step all s(0) are set to 1 with the exception of the root node which is set to 0. Convergence occurs according to a predefined criterion. s_i(t+1) representing the value of node i after (t+1) iteration steps and s_j(t) is the value of node j at iteration step t. If a directed edge from node i to node j exists, then a_ji= 1 otherwise 0. k_j^out is the outdegree of node j.

  (Equation 1)

 

The root node is no actual participant in the network, the root node is just used for convergence and thus the value that it accumulated until convergence t_c will then be split evenly among all other nodes as seen in equation 2.

(Equation 2)

 

Accounts posting the most influential tweets – judged by their respective like and retweet count – as well as the accounts with the most followers in total are reflected in the LeaderRank results. Political parties like “Bündnis 90/Die Grünen” and “CDU” seem to play a significant role. It is noteworthy that the former actually did not tweet with our exact phrases, yet due to their overarching emphasis on renewable energy and the environment, they were frequently mentioned, which led to a high LeaderRank. The high LeaderRank of the latter may stem from the previously mentioned presentation of their new climate agenda in the 38th calendar week. Furthermore, news organizations with a big followership like @-faznet, @-spiegelonline and @-welt also scored fairly high. Lastly, relevant actors linked to renewable energies as the federal minister for environment, nature conservation and nuclear security Svenja Schulze (@-svenjaschulze68) and the federal minister for economics and energy Peter Altmaier (@-peteraltmaier) are also found within the list of the ten highest LeaderRank scores.

The frequent occurrence of political parties, news organizations and political actors are relevant for public discussions but their presence and importance can also be anticipated a priori, thus their importance in determining actual opinion leaders that direct discussions and influence public opinions is marginal. Additionally, the search for opinion leaders is linked to identifying individuals and not organizations as political parties or news outlets. In contrast, individuals like Volker Quaschning and Luisa Neubauer become quite relevant as they are not relevant due to the information and beliefs as they share their own, independent perspectives, opinions and insights without being bound to an organisation. Their acceptance within the network is indicated through the high LeaderRank scores and their accumulation of likes and retweets. Thus, identifying and analyzing individuals that share similar characteristics seems to be a good way to actually determine the most influential individuals and at the same time the opinions that are dominant within the network. However, these individuals are underrepresented in the top 10 LeaderRank scores (Table 2).

The version of LeaderRank used did not integrate additional metrics like the like or retweet count. Still, these metrics are an inherent aspect of Twitter that influences how far a tweet spreads and how it is perceived by other users. Therefore, an adjustment of LeaderRank integrating like and retweet count but also the tweet frequency was used to compare the results (Table 3).

Using the Adjusted Leader Rank, accounts of political parties or news outlets like @-faznet, @-welt, @-spiegelonline, @-die_gruenen and @-cdu were downranked and with the exception of @-cdu do not even occur in the top 10 anymore (Tables 2, 3). Instead, an evident increase of individuals such as Karl Lauterbach (@-karl_lauterbach), Claudia Kemfert (@-ckemfert) and Simone Peter (@-peter_simone) can be observed in the top 10 which aligns with the concept of an opinion leader actually being an individual sharing influential ideas and thoughts in a specific niche and not an organization dealing with diverse topics.

6  Conclusion

The results of the various analyses show that the follower count alone is not performing well as an indicator for opinion leadership since the followers may stem from a domain not related to the topic at hand. Furthermore, the follower count cannot be reliably evaluated in retrospective since the follower count numbers were collected in 2020 while the network itself took place in 2019. Hence, it is likely that, in many cases, a big discrepancy exists between the collected and the actual numbers. Evaluating how well received tweets are within a network seems to be more reliable, yet it is not entirely reliable as likes and retweets could also originate from somewhere else. The individuals @-vquaschning and @-luisamneubauer both posted relevant tweets that received positive feedback and they both play a fundamental role in the network which is evaluated by LeaderRank, thus there is a high likelihood that they are indeed opinion leaders (Table 3). The combination of Twitter metrics (likes, retweets, post frequency) combined with LeaderRank also identified Svenja Schulze (@-svenjaschulze68), Simone Peter (@-peter_simone) and Karl Lauterbach (@-karl_lauterbach) as opinion leaders with a political background in addition to Claudia Kemfert (@-ckemfert) who has a background in science and thus can only be used to draw conclusions about the German Twitter community or rather the German speaking landscape of renewable energies.

Lastly, the present study does not yet provide information about the way tweets affect recipients, i. e. whether they influence the opinions of other people in the respective network. Further research is needed on this question, especially with regard to the perception of the messages by the recipients. Determining a change in opinion or behavior in the individual users based on specific tweets would require additional research into how different user groups perceive these tweets.