Red-tagging is an issue that hits close to home for any UP student. Chad Booc, a graduate of our department and an alumnus of UP CURSOR, is a famous victim of this issue. The circumstances surrounding his death are particularly polarizing. While many condemned the state for his death, quite a few lauded the government for its actions. This is an indication of the success of the state's red-tagging. Our frustration regarding this and the inherent dangers of red-tagging lead us to choose this issue as the focus of our CS 132 project.

Twitter has become a prominent platform for red-taggers to spread disinformation and target activists like Chad Booc. The ease with which information can be disseminated on Twitter means that disinformation can quickly gain traction and be amplified, leading to potentially severe consequences for targeted individuals and movements. These consequences include defamation, harassment, or even death, like in Chad Booc's case.

Through this project, we aim to be able to analyze the behavior and intent of red taggers on Twitter and identify their patterns of red-tagging. Through this, we can gain a better understanding on how red-taggers operate which can be helpful as a reference for combatting this kind of disinformation in the future.

What kind of behaviour did red-taggers exhibit in 2022 concerning Chad Booc and his activities in relation with:

1. frequency of tweets with mis/disinformation, and
2. the content of these tweets?

1. The frequency of tweets red-tagging Chad Booc and his activities significantly increased after his death.
2. The tweets containing dis-/misinformation predominantly talk about Chad Booc's death.

1. There is no significant increase in the frequency of tweets red-tagging Chad Booc and his activities before and after his death.
2. There is an equal distribution of topic clusters regarding Chad Booc's death; no specific topic cluster is significantly more predominant than others.

Analyse the frequency of tweets and usage of relevant keywords to identify red-taggers' behaviours.

We gathered tweets that red-tagged Chad Booc and Lumad schools. The scope of our data collection is 2022, the year he was killed.

In scraping for tweets, we used two sets of keywords, one for Chad Booc and one for Lumad school. Specifically, these are:

1. "chad booc (cpp OR npa OR terorista)"
2. "lumad (cpp OR npa OR terorista)"

With this set of keywords, we were able to collect a majority of tweets from 2022 that red-tagged Chad Booc and Lumad schools.

In order to speed up the data collection process, we used snscraper, a web scraper for social networking services (SNS).

Each row in our dataset contains one tweet. The columns we collected to describe each tweet are the following:

1. Tweet URL
2. Keywords Used
3. Account Handle
4. Account Display Name
5. Account Bio
6. Account Type
7. Account Join Date
8. Account Following Count
9. Account Followers Count
10. Account Location
11. Tweet Content
12. Tweet Type
13. Tweet Date Posted
14. Tweet Content Type
15. Tweet Likes
16. Tweet Replies
17. Tweet Retweets
18. Tweet Quote Tweets
19. Tweet Views

However, for the purposes of our analysis, we only used the following columns:

1. Tweet URL
2. Tweet Content
3. Tweet Date Posted

In the columns that we used, we did not encounter any missing values that needed filling. Similarly, we also do not have any columns that need normalising or standardisation.

The column Tweet Date Posted contained datetime values in the format MM/DD/YY hh:mm. In order to make it readable for code, it was parsed into a datetime object as an ISO8601 format string.

Due to the nature of our project objectives and data, we encountered no outliers that needed to be handled during preprocessing. Specifically, it is because these outliers are what we will be looking for in our time series analysis as peaks in the dataset.

In order to perform topic clustering on the tweets collected, the following basic procedures were performed:

1. Translate the content
2. Remove mentions, hashtags, links, punctuations, emojis, and extraneous whitespaces
3. Lowercase the content

This step was performed via Google App Script.

To get only important words and concepts in the tweets, a list of English stopwords is used to filter the words present in a tweet. This removes unnecessary words such as articles and pronouns in the content of the tweet. NLTK corpus' stopwords is used for this, which we extended to include additional words such as meaningless expressions. Then, NLTK's word_tokenize function is used to get the tokenised words to prepare for the next steps.

To get the common topics found across the gathered tweets, it is important that the content in each are in the same form. This is done by lemmatising the tokenised words using NLTK's WordNetLemmatizer. The help of NLTK's post_tag is employed to determine each word's part of speech for a more accurate lemmatisation.

In order to group the tweets by date, pandas' resample method is used with a frequency of 1 day.

1. February 25 - March 2

   Within 6 days there was an influx of tweets regarding Chad Booc's death

   Link1 Link2
2. March 3 - March 27

   There was a relatively low but stable number of tweets present for 25 days. Also, a peak was detected on March 12 when Chad's autopsy report was released.

   Link1 Link2
3. July 25 - August 3

   Another peak was recorded during the week of President Marcos' first State of the Nation Address.

   Link1 Link2

Using Latent Dirichlet Allocation (LDA), we were able to group the tweets into four clusters. The cluster of tweets predominantly talking about Chad Booc's death made up 27.46% of the total. Though this is one of the clusters that had the highest number of tweets, we were unable to find a statistically significant difference on the count in that cluster versus the others (χ2 = 1.96, p = 0.58).

1. Volunteers being red-tagged as NPA and Lumad schools being called NPA training grounds
Link
2. Makabayan block and other filipino activists getting killed and red-tagged
Link
3. Death of Chad Booc
Link
4. Chad Booc red-tagged as a member of the NPA
Link