‘Yorme, Yorme’: Modeling themes from COVID-19 concerns and public reports aired through Manila City government Twitter accounts.
How can public tweets help in making data-driven decisions for LGUs in times of a pandemic?
In this age of modern technology, it is, in a way, considered a waste if we do not harness its full potential to solve our problems. Big data is generated every single hour by social media platforms such as Twitter, Facebook, Instagram, etc. The current trend in almost all types of industries (universities, corporations, businesses, etc) is moving towards making data-driven decisions. This means that the core decision-making process depends on the collection of data for analysis of patterns to produce insights. And these insights, in turn, are used to know what actions should be done next toward achieving specific goals.
Twitter is a social platform where users can post anything and interact with anyone. In recent years, it has been used by large businesses to understand customer sentiments and how they can improve their products and services.
Since Twitter is widely used (330M monthly active users, around 10M in the Philippines [1]), it can be considered as a valuable tool to use to hear people’s sentiments on how the Philippine government, both national and local, can improve its services to the public. This is considered as a bottom-up approach since the data is directly from the voice of citizens.
In this article, we will discover major underlying ‘themes’ or topics present in tweets related to concerns and reports sent by people living in Manila City in this time of the COVID-19 pandemic.
We believe that the results of this research article will help the local government of Manila City in planning their next steps and improvement of their services (distribution of goods and aids, maintaining security and distancing measures, etc.) towards helping in flattening the curve of the current pandemic situation.
The Collection
Colloquially and informally known as ‘Yorme’, Manila City Mayor, Isko Moreno (@IskoMoreno) maintains an active Twitter account where he often posts daily activities to keep people updated on his government duties. His counterpart account, Manila Public Information Office (@ManilaPIO), serves as his official information disseminator account.
We managed to collect 7,405 public Twitter replies from March 1 to April 6, 2020, sent by locals of Manila City to the two mentioned official Manila City government Twitter accounts.
The photo below shows an example of tweets related to COVID-related concerns such as (a) reports on the improper distribution of relief goods, food packs, and aid, (b) praise for the efforts of the government officials, (c) current situation of the needy.
Plotting the date and time metadata that came in with each tweet, we see a steady growth of replies starting from the implementation of the Luzon lockdown last March 15, 2020. From then on, Manila City residents have relied on the Twitter accounts of Mayor Isko Moreno and Manila PIO in airing their concerns on the current situation. We can see a pattern on the spikes of tweet replies sent for every event that has transpired within the time period of the Enhanced Community Quarantine.
The Experiment
Topic modeling is a Natural Language Processing (NLP) method for automatic discovery of abstract and hidden topics or themes present in a large body of texts. This method is probability-based and each topic extracted from documents have their own corresponding probabilities that provide an explicit representation of a document itself [2].
In plain English, performing thematic analysis on a given corpus or body of text, in this case, thousands of tweet replies on COVID concerns, will allow us to identify the main ideas or subjects present. These ideas can then be used to discuss actionable initiatives that can be done based on the results of the experiment.
In this study, we performed biterm topic modeling [3] to the collected data. Biterm topic modeling is a variation of topic modeling based on word co-occurrence that is effective for modeling short texts such as tweets. You can find the published paper of biterm topic modeling here. We also performed bigram analysis of words for each topic to identify what words were frequently co-occurring with each other.
The Results
From the data, we were able to extract and identify four major themes from the tweet replies related to concerns and reports by Manila City locals. Each topic is analyzed and discussed thoroughly below.
Theme 1: The Needs of the Needy
The first topic to surface from the extraction is no surprise. It is the continuous plea of obtaining relief goods and food packs from the local government. The supporting topic words for this topic are shown below. We see supporting verbs or action words such as ‘tulong’, ‘mabigyan’, and ‘natanggap’. We also see some entities that are responsible for these jobs (‘chairman’) as well as the ones that we should prioritize (‘senior’).
Extracting the exact Twitter replies from Mayor Isko’s account, we see the consistent trend of locals reporting the following scenarios:
1. Taking too long to distribute relief packs.
2. There are certain qualifications regarding getting these goods or selective distribution (only registered voters will get the goods as reported).
3. Suspicions of hoarding of goods by officials (such as the barangay chairman).
4. There are stranded non-Manileños in Manila not being able to get goods due to the prioritization of Manileños.
Bigram analysis of some of the supporting topic words provides us with neighboring words commonly co-occurring with the specified word. In the diagram above, we see co-occurring words connected to selected words such as ‘relief’, ‘pagkain’, and ‘mabigyan’. This will give us a concrete idea on what the LGU of Manila should target for the next relief good distribution such as the following:
- The word ‘pagkain’ also occurs with the word ‘utang’. This may entail that one of the primary sources of food of locals is based on loans.
- The word ‘mabigyan’ which in this context means getting relief goods, co-occurs with entity words such as ‘janitor’, ‘frontliners’. This may mean a suggestive motion of prioritizing goods for front liners and utility persons.
Theme 2: When the Cat’s Away, the Mice Will Play
The second theme present from the collected data points to the local entities employed by the government to serve. Supporting topic words that describe this theme are ‘chairman’ and ‘pulis’. Aside from that, we can see some descriptive topic words that can be associated with them such as ‘gago’ (asshole), ‘racist’, and ‘pili’ (selective). These words may give us an idea of how officials such as our barangay chairmen and police officers have their way according to local reports directed to Mayor Isko.
Some words such as ‘pasaway’ (troublesome) and ‘lumalabag’ (violating) have an ambiguous context. Which entities were being troublesome and violating the lockdown rules? Is it the locals or the officials themselves?
Extracting a few Twitter replies based on the supporting topic words, we might have an answer. We see reports of locals directed to Mayor Isko on barangay officials and police doing the following:
- Negligence of duties and responsibilities in their respective barangays.
- Violating the liquor ban during the lockdown. Conduct of inuman near barangay halls.
- Cases of racism and not following lockdown rules.
Performing bigram analysis shows us even deeper and almost upsetting results. The following can be inferred from the result of the bigram analysis of co-occurring words for the current topic:
- The entity ‘police’ is associated with descriptive words such as ‘hayop’, ‘pang-aabuso’, ‘abusadong’, and ‘namalo’. Ideally, these words should not be associated with our policemen but the data from locals show otherwise. Are policemen allowed to abuse and spank (palo) people?
- The entity ‘tanod’ or a barangay law enforcer is associated with a peculiar word: ‘blumentrittkinakando’ or ‘bluementritt’ + ‘kinakandado’ (locked up). We might have accidentally removed the space between the two words. Nonetheless, this entails that some barangay tanods are reported to be locking up those who violated lockdown rules. Where? At Blumentritt Road in Manila.
- Lastly, the entity ‘chairman’ is associated with words such as ‘relief’, ‘bigay’, ‘katiwala’, and ‘galit’. Since the barangay chairman is the one heading the distribution of relief goods and quarantine passes, locals are expected to contact him/her. With the word ‘galit’ (angry) associated, does this mean that some barangay chairmen tend to outrage in frustration during these times?
Theme 3: Vico and the Tricycles
The third theme present from the data collected is not entirely a Manileño, to say the least. Welcome, Mayor Vico Sotto (@VicoSotto), the esteemed Mayor of Pasig City. To give context, Mayor Vico was denied his request by Malacañang of using tricycles for mobilization of health workers needed in some areas of Pasig due to social distancing issues. Mayor Isko, on the other hand, was granted in utilizing e-tricycles for the exact, same reasons. Locals are quick to point this out based on replied under Mayor Isko’s Twitter account. We see supporting topic words such as ‘sotto’, ‘dilg’, ‘etrike’, ‘tricycle’, ‘workers’, ‘pasig’, and ‘gaya’ (copy).
Based on the actual tweets containing the supporting topic words, we see reports of the following:
- Locals siding on either party. Some siding with Mayor Isko for following DILG protocols, and some siding with Mayor Vico for fairness.
- Plea on using tricycle as a mode of transportation for specific vicinities of Manila to transport sick people and health workers.
- Tricycles parked where they are not supposed to be and may cause disruption. Places are reported such as Quinta Market.
Performing bigram analysis shows us a closer look on the topic at hand. The following can be inferred from the result of the bigram analysis of co-occurring words:
- People brought the issue of the ban of tricycles for emergency and mobilization purposes in Pasig City at Mayor Isko’s account. Citing why he was able to use e-tricycles while Mayor Vico was not.
- The request of using other modes of transportation such as jeepneys so that drivers may maintain their income daily.
- Suspicion of the government being biased to Mayor Isko. Words such as ‘pinaboran’, ‘aprubado’, and ‘pinayagan’ may support this claim.
Theme 4: The Learners’ Dilemma
Lastly, we have the fourth topic and maybe one of the most crucial ones to focus on. Students all over universities in Manila have taken their concerns and pleas on the immediate transfer of traditional learning to online learning (e-learning) to Twitter, especially on Mayor Isko’s page. From the supporting topic words below, we see concerns such as ‘wifi’, ‘connection’, ‘suspend’, ‘class’, ‘temporary’, and ‘school’ connected with the entities that will be detriment from such as ‘estudyante’ or ‘students’.
Looking up the exact tweets using the topic words, we see reports of the following:
- Students don’t have a stable internet connection to continue and attend online classrooms.
- Students are afraid to go out and risk their lives just to go to internet cafes and attend online classes.
- Request for higher governing bodies concerned (CHED, Malacañang) to suspend online classes nationwide due to connection problems.
- Students cannot afford expensive mobile data plans.
Performing bigram analysis shows us a closer look on the students’ concerns regarding the implementation of online classes of universities. The following can be inferred from the result of the bigram analysis of co-occurring words:
- Plea for consideration of students with no internet connection ( ‘cases’, ‘online’, ‘better’, ‘capable’).
- Students have a hard time adapting and performing assessments in an online learning environment (‘makapagexam’, ‘exams’, ‘video’, ‘pinapagawa’).
- Waiting aimlessly for an appropriate action or suspension from universities given dire situation ( ‘appropriate’, ‘isipin’, ‘cases’, ‘could’, ‘suspend’, ‘muna’).
Once we recover from this pandemic situation, it might be best to suggest to universities to formulate an e-learning plan for careful integration of online learning modules so that students will not experience a door-to-the-face scenario where they are forced to transfer to a different learning environment and have a hard time in adapting.
Conclusion
As we progress with our days while waiting for an immediate vaccine or cure for the pandemic, our best effort is to flatten the curve or to slow down the transmission of COVID-19. But in order to maximize our efforts, we have to follow the rules and strive for continuously improving our services to the people. May it be in the government, industry, or university.
The purpose of this research is to help the local government of Manila City make data-driven decisions to further improve its services to the public by analyzing information provided by its constituents, the people of Manila. Obtaining the sentiments, concerns, and reports sent by locals themselves is one the best way to do this as it follows the bottom-up approach. Using NLP techniques such as topic modeling and bigram analysis, we were able to automatically extract underlying themes present in the data without manually reading each tweet one by one.
The future direction of this study is presenting the results to concerned government officials and help them re-align their decisions in the upcoming days as well as create new ones based on the insights provided for each topic.
Author’s Note
If you have further questions and clarifications with this study, feel free to contact me. Don’t forget to clap! :)
By the way, I am Joseph Imperial, instructor and NLP researcher from National University. Most of my research areas are in Natural Language Processing and Machine Learning.
You can also visit my profile here: https://sites.google.com/view/josephmarvinimperial/
References:
[1] 10 Twitter Statistics Every Marketer Should Know in 2020. Oberlo. https://www.oberlo.com/blog/twitter-statistics
[2] Blei, D. M., Ng, A. Y., & Jordan, M. I. (2003). Latent dirichlet allocation. Journal of machine Learning research, 3(Jan), 993–1022.
[3] Cheng, X., Yan, X., Lan, Y., & Guo, J. (2014). Btm: Topic modeling over short texts. IEEE Transactions on Knowledge and Data Engineering, 26(12), 2928–2941.
