2022 Philippine Elections

[ANALYSIS] The Philippine cyber-political divide

Gilbert Chua
[ANALYSIS] The Philippine cyber-political divide
Political polarization is observed on Twitter amid the 2022 Philippine presidential elections

“The Philippine cyber-political divide: Political polarization on Twitter amid 2022 Philippine presidential elections” is a study presented by the author in the second #FactsFirstPH research briefing held on March 18, 2022. The full copy of the research is reposted with permission.

The spread of information (including disinformation) has been extensively studied, especially given the proliferation of “fake news” and how they have since impacted election outcomes. 

[ANALYSIS] The Philippine cyber-political divide

For example, the recent work of Marc Jones published in the International Journal of Communication titled, “The Gulf Information War| Propaganda, Fake News, and Fake Trends: The Weaponization of Twitter Bots in the Gulf Crisis,” revealed how Twitter “bots were used to manipulate Twitter trends, promote fake news, increase the ranking of anti-Qatar tweets from specific political figures […] amplifying propaganda discourses beyond regional and national news channels.”

In this study, we were interested in seeing how Twitter users interact with hashtags related to Philippine elections. Here, “interaction” looks beyond the use of the hashtags and also examines how users retweet or respond to a tweet containing hashtags of interest.

Specifically, we wanted to uncover levels of political polarization between supporters of two leading presidential candidates – Ferdinand Marcos Jr. and Leni Robredo – noting that political polarization has affected the susceptibility of groups to fake news.

In the United States, studies have suggested that social media users who report hating political opponents are – according to political science researchers from Aarhus University – also the ones “most likely to share political fake news and selectively share content that is useful for derogating these opponents.” 

These findings agree with the results of two studies done in 2019 with the first one showing that the sharing of fake news was most prevalent among conservative voters and the second one reporting that pro-Trump supporters were more likely to share fake news. These are important and very relevant studies as they were conducted under the lens of fake news in the context of national elections – specifically the election of a president.

We studied Twitter data containing nearly 450,000 tweets with topics (phrases or hashtags) that trended from October 1, 2021, to March 12, 2022. Trending topics were chosen from the 10 highest rated topics or hashtags (on Twitter) in the Philippines for every hour of the same period. These topics were made visible to Twitter users, highlighting them on each user’s “trending” list.

Manual annotation

Philippine election-related topics were manually chosen by sorting through 8,138 unique trending topics from October 1, 2021, to March 12, 2022. The main criterion for choosing these election-related topics was their mention of a candidate’s name, color, or event featuring presidential candidates. Through this process, 128 topics were identified and assigned to their corresponding candidate/s.

Table 1. Sample table of hashtags attributed to corresponding candidates. A total of 128 hashtags/topics were identified and attributed to each candidate along with their polarity. In the case of presidential candidates Ping Lacson, Isko Moreno, and Manny Pacquiao, each of them has only a single topic associated with them – their names – and these mentions are assumed to be positive in nature as there is no additional reliable information available.

In addition to assigning topics to candidates, each candidate-related hashtag was also given a polarity of being either positive or negative, depending on whether the hashtag was promoting the candidate or attacking them, respectively. 

Based on the polarity of topics, two parties exhibited polarization explicitly – the parties of Vice President Leni Robredo and of ex-senator Ferdinand Marcos Jr. It is the accounts of those that tweeted the most about these candidates that this study focused on. 

Table 2. Breakdown of tweets attributed either positively or negatively to presidential candidates and/or their vice presidential partners. We see that more than 3/4 of the tweets are positively attributed to Leni Robredo and running mate Kiko Pangilinan. Isko Moreno is the second most tweeted about candidate.

Support classification

To estimate each account’s level of support for a candidate, we identified the net number of positive tweets it made for each candidate. If an account tweeted about the topics “#Kakampink” 5 times (positive for Robredo), “#LeniDuwag” 3 times (negative for Robredo), and “#LabanMarcos” and “#BringBackMarcos” a total of 5 times (positive for Marcos), then that account was tagged as being a Ferdinand Marcos Jr.-oriented (net_mjr) account since it had shown the greatest interest toward Marcos. From these, we were able to identify the accounts that were net_leni, net_isko, net_mjr, net_ping, and net_pacquiao.

Table 3. A weakness of this method is that mentioning a candidate does not necessarily always connote support for him or her. Applying the oft-quoted adage, “there’s no such thing as bad publicity,” we attributed all mentions of a candidate to be positive in the absence of obvious context in a topic. We revisited this assumption later and described a method of how we can recontextualize mentions of a candidate based on the behaviors of accounts they associate themselves with.

Negativity

A way to observe polarization is to measure the amount of negativity between one group of people with another. We found that accounts that tweeted largely about Leni Robredo had almost thrice the number of negative tweets compared to accounts that tweeted mostly about Ferdinand Marcos Jr. 

Controlling for the total number of tweets, we found that the proportion of negative tweets by net_mjr accounts was 6 times higher relative to tweets by net_leni accounts. This suggests greater negativity associated with the tweets by net_mjr accounts compared to net_leni accounts.

There was also greater participation among net_mjr accounts with negativity. There was almost 3 times more participation in negativity net_mjr as measured by the number of unique accounts that tweeted about a negative topic at least once.

Table 4. Statistics of negative activity per candidate. Accounts that have tweeted more about Ferdinand Marcos Jr. tend to have greater proportions of negativity in terms of tweets and the accounts that have tweeted about a negative topic since October 1, 2021.

Looking at Figure 1, we found the greatest changes to be in the topics mentioning Ferdinand Marcos Jr. negatively and those that mentioned Leni Robredo among all users since January 1, 2022. 

Breaking down the trends by the account’s orientation, we saw that net_leni accounts were the ones largely responsible for the negative tweets related to Ferdinand Marcos Jr. We also noticed a very sharp uptrend in the number of tweets by net_leni accounts starting March 3, 2022.

Figure 1. Snapshot of the trends attributed to each candidate with respect to values on January 1, 2022. Overlapping trends show negative activity related to Ferdinand Marcos Jr. and Leni Robredo.
Figure 2. Snapshot of the trends attributed to each candidate with respect to values on January 1, 2022, for net_leni accounts. We find the greatest changes to be in the topics mentioning Ferdinand Marcos, Jr. negatively and those that mention Leni Robredo’s name.
Figure 3. Snapshot of the trends attributed to each candidate with respect to values on January 1, 2022, for net_mjr accounts. We find the greatest changes to be in the topics mentioning Ferdinand Marcos Jr. and those that mention Leni Robredo coinciding with that of net_leni accounts.

It is interesting to note that based on the trends for net_mjr accounts, there was a noticeable increase in the number of their tweets that mentioned Leni Robredo. 

This highlights one of the limitations of the tagging methodology as it is likely that the tweets from accounts oriented towards Ferdinand Marcos Jr. and which mentioned Leni Robredo were truly negative instead of being neutral or positive. 

It is also worth highlighting that this increase in the number of tweets about Leni Robredo by net_mjr accounts coincided with the increase of tweets by the net_leni accounts that mentioned Leni Robredo. In the same time frame, the number of tweets mentioning Ferdinand Marcos Jr. also increased, but not nearly as high as the number of tweets by the net_mjr accounts about Leni Robredo.

Retweet network

Polarization can be further highlighted by observing the interactions between the opposing factions. To monitor these interactions between multiple accounts in these groups, we compared their retweets within their network.

An account was considered part of this network when it either quoted or had been quoted by another account, resulting in a tweet or message becoming part of the top 10 trending topics.

This quoting behavior captured the spread of content between accounts as new accounts built on existing tweets and either passed on or modified their message.

Table 5. Number of accounts associated with each candidate that have quoted a tweet and wrote text that is among the top trending topics. Leni Robredo has the largest proportion followed by Ferdinand Marcos Jr.

We found that an overwhelming majority of public accounts that quoted users and tweeted about trending election-related topics tweeted about Leni Robredo more than other candidates. This can be seen in Figure 4 where there is an exceptionally large number of pink accounts in the middle. However, we can clearly see the presence of another cluster of users that quote one another. 

Red accounts that tweeted positively about Ferdinand Marcos Jr. can be found in the cluster on the right. What is interesting to note is that within the secondary cluster, there were many accounts that still tweeted about Leni Robredo – attributed to those simply mentioning the topic “Leni” rather than any of the other positive hashtags related to Leni Robredo. 

This shows that those who discussed the topic “Leni” and did not use positive hashtags for Leni Robredo tended to quote and be quoted by net_mjr supporters. To examine echo chamber-like behavior, we looked at each set of accounts corresponding to net_leni and net_mjr accounts.

Figure 4. Quote network colored by their net support of a candidate. The largest cluster colored pink corresponds to those that tweet largely about Leni Robredo. To its right, there is a secondary cluster colored red corresponding to accounts that tweet largely about Ferdinand Marcos Jr.

We found that net_leni supporters had a very strong tendency to quote and be quoted by one another. More than 95% of those who quoted net_leni accounts were also net_leni accounts. Following a similar behavior, 87% of all quotes by net_leni accounts were of tweets by other net_leni accounts. This behavior suggests a strong tendency of net_leni accounts to quote and build on one another’s tweets almost on an exclusive basis.

On the other hand, net_mjr accounts exhibited a different behavior from net_leni accounts. Only 50% of those who quoted net_mjr accounts were themselves net_mjr accounts, with net_leni accounts making up 48% of those who quoted them.

Interestingly, 49% of quotes by net_mjr can be attributed to net_leni accounts while only quoting other net_mjr accounts 27% of the time. This suggests a greater diversity in the topics that net_mjr accounts quoted, as almost half their quotes came from outside their faction. This suggests that net_mjr accounts engage others outside of their groups more relative to net_leni accounts. This diversity in engagement is of particular interest as it suggests a fundamental difference in the behavior of the two factions.

Figure 5. Quote network colored by the identified communities. The largest cluster colored pink corresponds to those that tweet largely about Leni Robredo. To its right, there is a secondary cluster colored red corresponding to accounts that tweet largely about Ferdinand Marcos Jr. Within the secondary cluster, there are many accounts that still tweet largely about Leni Robredo.

To better highlight these clusters of supporter factions, we applied a community detection algorithm to identify those that roughly belonged to the same group, based on their tendency to quote and be quoted by each other. Groups of accounts that tended to quote and be quoted by other accounts had the same color assigned to them.

As seen in Figure 5, the algorithm was able to identify the single large cluster in the middle corresponding largely to net_leni accounts shown in Figure 4. The secondary cluster in red similarly captured all the net_mjr accounts and even captured net_leni accounts that frequently interacted with them but did not use other positive hashtags related to her.

Inspection of ~2,500 accounts grouped into the red cluster and tagged as net_leni, showed there were accounts that tweeted only negatively about Leni Robredo without tweeting about topics related to other candidates. This means that some net_leni accounts are actually supporters of Ferdinand Marcos Jr. yet they tweet more negative things about Leni Robredo than positive things about their own candidate. The activity of net_mjr accounts tends to be made of two big components – negative campaigning against Leni and positive tweeting about Ferdinand Marcos Jr.

These results imply the need for more sophisticated ways to analyze tweets – specifically looking at how these accounts interact with one another. Methodologies that will analyze the polarity of individual tweets relative to personalities need to be refined further for us to better understand the dynamics at play in cyber, political, and social spaces among Filipinos.

Conclusion

Previous studies in the United States have shown that political polarization plays a critical role in the spread of “fake news” and the proliferation of disinformation. 

Our analysis of the behaviors of almost 130,000 accounts engaged in Philippine election-related conversations highlights polarization between supporters of two presidential candidates – Ferdinand Marcos Jr. and Leni Robredo. 

We found that each faction varies in both number and behavior. In terms of tweeting about negative topics, those that tweet mostly about Ferdinand Marcos Jr. tend to exhibit more negativity. Accounts that tweet about Leni Robredo mostly tend to quote and be quoted almost exclusively by others like them about the same topics. 

For accounts that tweet about Ferdinand Marcos Jr., however, interactions can be broken down mainly into two – interaction with other accounts within their network and interaction that involves quoting from accounts that tweet almost only negatively about Leni Robredo. 

Further research is needed to improve our understanding of all these interactions and pin down how disinformation spreads in this type of a politically polarized environment in the Philippines. – Rappler.com

Gilbert Chua is a PhD in Data Science Student in the Asian Institute of Management and an honorary member of the Analytics, Computing, and Complex Systems laboratory (ACCeSs@AIM). He is interested in delving deep into the workings of social systems through the lens of machine learning, graph theory, and complexity.