1 of 19

Visualizing Tweets with

Positive and Negative Sentiment

Towards Cities Affected

by Natural Disasters

in Metro Manila

Alfonso Secuya, Jared Rivera

2 of 19

Motivation:

​

Natural disasters cause different attitudes

within a single location

3 of 19

The Disaster Tweets Dataset:

  • scraped from Twitter using TWINT, a Twitter scraping tool
  • Tweets scraped had to have at least:
    • One Metro Manila City Keyword (ex: “pasig”)
    • One Natural Disaster Keyword Related to Rain or Fire (ex: “ulan”, “sunog”)

4 of 19

The Disaster Tweets Dataset:

  • Has a total of 24,331 tweets by 14,147 Twitter users
  • Each entry contains the following information:
    • Date and Time of the Tweet (created_at)
    • Text content of the Tweet (tweet)

5 of 19

Scope of Analysis:

​

1) Sentiment for the study is solely based on emojis and emoticons

6 of 19

Scope of Analysis:

​

2) Study will only focus on Metro Manila cities

7 of 19

Scope of Analysis:

​

3) Study will only focus on natural disasters related to effects of rains and fires

8 of 19

Scope of Analysis:

​

4) Study will only focus on tweets created

within February-August 2019

9 of 19

Cleaning and Preprocessing

  • Text content was converted to lowercase.
  • All special characters were removed, except for emojis and face emoticons.
  • A city column was created based on the presence of a city keyword within the tweet. (“las pinas” and “las piñas” -> “las piñas”)
  • Tweets containing 2 or 3 cities where duplicated and assigned different values for the city column.

​

10 of 19

Cleaning and Preprocessing

  • A sentiment column was created which contained “positive” and “negative” values
  • The sentiment value was assigned based on the majority of positive or negative emojis and emoticons present in a tweet
  • Positive and negative emojis and emoticons were based off Kralj et. al’s (2015) work
  • A total of 8,234 entries remained

11 of 19

Data

12 of 19

13 of 19

Visualizations

  • Majority of the tweets are from Manila.
  • Number of tweets may be from students.
  • There is a significant difference in positive and negative tweets overall.

Number of Tweets in Cities in Metro Manila

14 of 19

Visualizations

Word Cloud of Vocabulary from

Positive Tweets

15 of 19

Visualizations

Word Cloud of Vocabulary from

Negative Tweets

16 of 19

Visualizations

  • Most tweets are found around August and September.
  • There is a big difference in negative tweets from March to August
  • There is also a significant difference in positive tweets from May to September

17 of 19

Visualizations

Average

  • There are more tweets occuring later at night.
  • A sudden increase in tweets occuring around 5AM and 6AM.
  • The average number of positive tweets and negative tweets are consistent.

18 of 19

Conclusion

  • Most attitude towards Natural Disasters are negative regardless of location.
  • Maybe from students tweeting nights with heavy rains.
  • Also maybe from students tweeting the morning of heavy rains.
  • This analysis can help government officials understand “overall” sentiment towards natural disasters in Metro Manila.

19 of 19

Limitations

  • A bigger dataset would results in more insights.
    • Can have more than 1 year so we can see the yearly trend of tweets.
  • More accurate sentiment can better be analyzed through text than Emojis and Emoticons use only
    • Ex: Sarcasm, Sentence Structure
  • Better techniques are needed to differentiate sentiment between two cities present in one tweet