Data Journalism on Twitter - 2020

This research observes users and communities around the #ddj hashtag on Twitter. It intends to find the most important ones and to explore the geographical factors associated with them. It aims to understand the online ecosystem of data journalism and the real people behind it.

In general, the accounts that are actively engaging in data journalism discussion in 2020 have formed several groups based on their roles in the news industry, the media they are involved in, and the geographic area they locate. Most of the narrative readings resonate with the values and themes associated with news and investigative journalism at different levels (Bounegru et al., 2017).

This research combines qualitative and quantitative approaches. It is composed of three stages: the data preprocessing, data mining and data interpretation. Only users with the most in-coming or out-coming mentions were kept; the Leiden algorithm generated well-connected clusters. These users’ Twitter profile information was then analysed, trying to map their network and explore the characteristics of each cluster.

Using the Leiden algorithm, twelve communities were extracted: two of them were pruned since they were too small and sparse; the others were examined qualitatively to discern their characteristics.

Mapping the communities in the online discussion around data journalism hashtags and finding out characteristics of different clusters.

The communities were generated by analysing the network of mentions between users, that is why some clusters seem to exist around one specific account. These clusters are generally associated with the partnership of organizations or with geographical factors.

This section studies the individual contributions on the #ddj environment.

Finding the most active users and the most influential ones, try to understand if they correspond and if they belong to any community, or if any lonely wolf exists.

The analysis was taken following two approaches and comparing them.

Local approach

This approach only considered people belonging to a community, to understand who, among the most connected users, emerges.

Global approach

This approach considered the whole dataset to verify if the local rankings corresponded to the global ones.

There are three classes of users: the mostly inactive, the spreaders (those who mention other users), and the ones that are mostly mentioned: the stars.

As we can expect, the majority of users fall in the first category: the casual users. Most of them were excluded from the analysis. For what concerns the other two classes, it emerged that they don't often overlap: the cases at the limit of the distributions tend to primarly mention other users or to be cited by many people.

This is not surprising since there exist accounts made for spreading specific contents made by and addressed to a precise target, in this case (data) journalists, activists, and designers.

It is useful to understand in what measure each user contributes to the Data Journalism community on Twitter. This subsection reveals the most active or influential users, and the spreaders. This classification follows the first approach, excluding the users outside the communities.

A profile was considered active if it produces a large amount of tweets, influential if many other users cite its work. It was named a spreader if it mainly retweet other users' work.

Two rankings were realised: one considering retweets, and the other excluding them, so prioritising original material.

At a first glance, it seems more useful to look at the data without retweets. Here we can find that the 4 out of 10 users belong to cluster 1. It is not so surprising, considering that it is primarly populated by big news agencies.

@gijn can be considered the most active user, also considering the other plot: as we see, it retweeted only 22 times, much less than the 192 of @pinadrag. These two users are significantly over the average number of #ddj tweets in 2020, around 2.5 (excluding the zero values).

This second analysis focuses on the level of interaction between the users (excluding retweets). The situation displayed by the plots shows cluster 1 as very present among the top 10 spreaders.

While @pinadrag disappears, @gijn continues to dominate the rankings, consolidating its role of authority in Data Journalism, at least on Twitter. Also @datajournalism, the third most active user, confirms its position of importance becoming the second most mentioned in the network.

Another aspect to notice is that cluster 0 receives a large amount of mentions. Given the nature of its composition it can be foreseen that these are principally interactions inside the cluster. A qualitative proof of that is the visual display of the community in the network visualisation at the beginning of the article.

Global placements (inside the network) are fundamental to see local rankings in perspective, because important actors not strongly socially connected could emerge.

The global plots show the presence of a consistent number of independent users at the top of the placement. These could suggest the presence of less represented communities, like the Russian one (suggested by @gijnRu) and the Turkish one (suggested by @DagmedyaVeri).

An anomaly was found: the two users at the top of the ranks, @EscolaDeDados and @journalismoDados probably represented the same institution. The latter was deleted before March 2021 in favor of @EscolaDeDados (cluster 0). This study, then, excluded @journalismoDados because it imbalanced the data without adding any contribution. In any case, it can be assumed that @EscolaDeDados has a quite important role, even if its first placement could be due to its strong relation with the twin account.

Again, @gijn, @pinadrag and @datajournalism appear to be often mentioned, confirming the local observations.

Conclusion

This analysis unveiled the presence of absolutely important users, like @gijn and @datajournalism, and uncovered that the most mentioned users, also in a global context, belong to a community. It also confirmed that the amount of tweets is not the only key of popularity: the global analysis of activeness shows many active users disappearing in the rankings of mentions; this suggests the importance for an individual to be part of a community, and at the same time the difficulty of small clusters to emerge in the Twitter environment.

The analysis explores the geographical distribution of the data journalism community. In addition, it will analyse the factors that may influence the formation of communities on Twitter in Europe, the US, Nigeria and Brazil, and the reasons why some regions are missing from this space.

What is the geography of these voices, and who is missing from this space.

The accounts located in the European countries and in the US constitute the majority of the 197 users in the communities and have a relatively close connection with users inside and outside their cluster. Also, the most mentioned ones and the top mentioners are mainly from these regions, suggesting the importance of European and American organizations in the data journalism Twitter environment.

In parallel, the accounts from other parts of the world, like Nigeria and Brazil, tend to interact within their cluster, except for some local organisations being mentioned by international accounts.

The absence of voices from most of the countries on other continents may suggest that these countries have not built up a Data Journalism system of international relevance.

Missing voices

While major western media such as the BBC and New York Times build their Data Journalism teams with cutting-edge methods, most of the Asian accounts are either posting content that has low connection with data journalism or limited in number. Japan and China are taken as two examples to illustrate why some countries may be considered the missing voices in this research.

This research strongly depends on two factors: the chronological factor and the data source. To begin with, 2020 saw two important events that could have distorted the analysis: the start of the pandemic and the US presidential election; it is possible that the discussion topics determined in some measure the appearance or not of some users. However, the top of the rankings don’t seem to be so much altered: the most important users actually constitute a reference for data journalists.

The dataset source could pose more problems, since some data journalism communities do not organize through the same hashtags, or do not publish on Twitter at all. This bias suggests that most of the ten communities extracted in analysis #00 could be considered part of a unique, bigger, occidental cluster, since the majority of accounts and communities come from Europe and the United States.

This imbalancing could be associated with local media freedom and access to information or to the usage of different means of communication. Since other datasets to measure these phenomena are quite difficult to produce, and that the local factors seem to be so influencing, a qualitative approach was preferred: looking at a few specific cases allowed to reflect on the global Data Journalism environment and how social media could influence its growth, limiting the contacts with communities that are less active in the western mainstream platforms.