This section discusses the results obtained, and more specifically, the questions and objectives presented in section 1.1. Also, the results collected are compared to other results in the existing academic literature, especially focusing on the studies with a focus on the Catalan and Spanish scenario.
First, we will focus the discussion on the presence of echo chambers in the global conversation on Twitter related to the #14F elections, and how they behave within themselves and how they interact with others:
Q1
Does the conversation around #14F reflect echo chambers for each political party community? How do these echo-chambered communities behave within themselves and how do they interact with others?
Yes, results obtained in section 4.2.1 clearly reflect the existence of echo chambers in the Twitter conversation related to the electoral campaign. All the 8 communities detected have more than 80% of interactions inside their own community, with a peak 94,26% of VOX community interactions inside their own community.
The tweet content analysis in section 4.4.3 reinforces this statement because it is possible to detect that the most used words are related to campaign slogans, asking for the vote, or referring to a relevant political figure of their community. Even though we have detected some words that possibly show inter-community attacks (“Illa”, “VOX” and PSC”), these are used as an internal product to attack the opposition, and not to engage in a real conversation with the opposite communities.
Regarding these echo-chambered communities, PSC and VOX are the most hermetic, with percentages so high that we can affirm that they almost only interact within their communities. At the same time, JxCat and PSC are the more prolific communities if we look at the average interaction rate per user (considering all community interactions). On the other hand, CUP community is the one that has less interaction in its own community and the one with the lowest average interactions per user. These results contribute to the topic knowledge, confirming what was found in Guerrero-Solé (2017) and Del Valle & Bravo (2018), but contradicting results obtained in Balcells et al. (2020) where blocks pro and against independence interacted with each other. The difference in the results obtained with the Balcells et al. (2020) study is related, in our opinion, to the fact that they focus on the replies, while this study has a majority of retweets inside the data collected.
Q2
Can the three main political blocks (independentist, constitutionalist and Spanish government parties) be grouped into three greater echo chambers?
Yes, the results of section 4.2.1 show that echo chambers are even stronger when we group these 3 political blocks, obtaining an astonishing result of more than 94% of interactions inside each block, which basically means that there is almost no conversation outside these bubbles. The highest inter-block community interaction is the “Spanish government” parties block which has around 3% of active interactions with the independentist block (mainly due to Comuns interactions with them). On the other side, we can observe the deep isolation between independentist and constitutionalist blocks, with only 0,39% and 0,66% of interactions respectively.
This analysis is reinforced by the sentiment analysis in section 4.4.3, where we can detect similar sentiment performance on each of the 3 blocks, being the independentist (pro-Catalan government) the more positive communities (above 38% positive sentiment tweets), while the constitutionalist communities are the more negative (more than 50% negative sentiment tweets), which can be related to the known fact that they used an aggressive campaign style against the independentist. Also, as seen in Table A11, these differences are very present when the language used is analyzed.
Again, these results reaffirm what was analyzed in Guerrero-Solé (2017) and Del Valle & Bravo (2018), but also contradict the analysis of Balcells et al. (2020) more clearly in this casuistic due to the affinity block union.
Q3
Are there any differences/similarities between the political affinities in the digital ecosystem and in the Parliament that resulted from the election?
As answered before, the similarities in the political affinities in the digital ecosystem compared to the real-world relations are substantial, especially the fact that the party communities can be joined in a political block, which matches for example the investiture agreement that allowed Pere Aragones to become President (ERC + JxCat + CUP). At the same time, we can observe that apart from the three blocks separation, Comuns can be relatively closer to the independentist block, which also matches the punctual agreements that they arrive at in real life.
Also, the study (Table 8) showed that the size of digital communities does not correlate with the results of the elections: PSC and ERC were the most voted parties, but are not the largest communities, and also VOX is the biggest community when it was placed in the fourth position. Observing the centrality of official accounts of the parties and relevant political figures, we have detected that only 2 out of 8 candidates are on the top 3 of their community users by in-degree.
Secondly, a complete analysis of social bot penetration on the global Twitter conversation has been performed. In relation to that, as in similar studies (Pastor-Galindo et al., 2020), it is not possible to completely affirm their existence or ensure that our classification is 100% accurate, but the methodology used matches previous methods (Stella et al., 2019; Varol et al., 2017, Pastor-Galindo et al., 2020), so this study adds these results to the global knowledge on the Twitter analysis on election periods. More concretely if we focus on the research questions presented before:
Q4
Were social bots massively used in the digital #14F electoral campaign?
This is a difficult question to answer with the data available, even though the widely recognized app Botometer has been used, and conservative parameters (percentile 95) have been considered, it is not possible to affirm that social bots have been massively used by a specific political party or community. But with the results obtained and the methodology used, we can state with a high probability that there are some results that certainly point to suspicious parameters in some of the communities.
Q5
Does any community have a disproportionate number of them?
The data available on the bot analysis by community points especially to the VOX community as the one with the highest numbers in nearly all the casuistics studied. First, the community data exploration shows a high rate of accounts created in 2020 and 2021, as well as numberplate accounts percentage, compared to the rest of the communities. As explained before, this could match the fact that it is the newest party with considerable growth in the last years, but when studied in depth by the bot analysis performed, it could match with the fact that social bots were created once the pandemic started, and digitalization became even more present in our societies. This result matches the results obtained in Pastor-Galindo et al. (2020), where the cumulative number of bots in VOX community during the analyzed period was much higher than other contenders.
When studying other communities, we can observe that some of them have outliers in some particular study case: Cs have a high percentage of bots inside their users pool, Comuns have a high percentage of their alleged social bots created in 2020, JxCat have an unnatural number of alleged social bots created in 2021 (85 new accounts for only 45 days until February 14th ), and also Comuns and PSC have a high percentage of “numberplate” social bot accounts.
But when studying VOX community, it is possible to notice that they have high numbers and percentages in nearly all of the cases under study: they have the second-highest percentage of bots compared to the community size, the lowest percentage of human users (even though we have seen that they have a considerable amount of users on the boundary limit between human and unclear), the highest number and percentage of bots created in 2020, and the second-highest number of numberplate bots by percentage. All this information allows us to confirm the strong presence of social bots in VOX community, which could be even enlarged if the limits on the bot tagging were less strict.
Q6
How do these bots interact with human users, in terms of interaction type (RT, reply or quote)?
When analyzing the interaction patterns by type of user, it is possible to observe that human users have what can be qualified as human behavior: they usually interact with other human users and in very few cases they interact with bot users. Also, human users have a considerable amount of quote and reply tweets, which fits in normal use of the Twitter application. On the other hand, bot users nearly always retweet (more than 96% of tweets captured), and half of these interactions go to human users. This allows us to state that the relevance of bots in the general conversation is low, and they are mostly used to amplify the messages thrown by human users. This casuistic has been noticed previously in Gonzalez-Bailon et al. (2020).
If observed in detail by communities, these interactions showed that, obviously, the more common interactions are the human-human relations, with the only noticeable result of the low percentage inside VOX community (probably due to the low human tagged user percentage), and the appearance of the human-human relation between bonded communities with affinities (inside independentist block). Focusing on the bot-human relations by community, there are relevant results regarding JxCat, Comuns and Cs communities, with a high number and percentage of relations of this type.
Q7
Are there any differences between social bots and human users regarding the text used and its sentiment analysis?
No considerable differences have been noted in the text analysis between bot and human users, as well as in the sentiment analysis of their respective texts. The probable reason behind that is the high rate of RT in bot tweets, which implies that bots are used to amplify the messages of human users, thus the text and sentiment analysis have a high similarity between both types of users. These results do not coincide with previous investigations such as Stella et al. (2018), where bots propagated primarily negative and inflammatory content.
Finally, some limitations have to be recognized during the development of this study. First, regarding the methodology and tools used (Botometer and Polyglot mainly), which are widely used tools and methodologies but could have also some biases or errors in their own operation. Secondly, a conceptual debate on the fine line that separates echo chambers from other phenomena such as homophily or confirmation bias, could generate controversy on the results and conclusions obtained. In addition, there is a considerable limitation due to the poor variety of tools to perform sentiment analysis in Spanish, and even worse in Catalan. This last fact greatly constrains the text sentiment analysis presented.