Analysis of Filipino Mood Swings within a Day using Tweets

Rodalyn A. Balajadia, Vincent Louie L. Maglambayan and Maria Teresa R. Pulido

Department of Physics, Mapúa University, Manila City, 1002, Philippines

Keywords: Data Analytics, Big Data Algorithm, Social Science and Implications for Big Data, APIs, User Evaluations

and Case Studies, Microblogging, Twitter, Opinion Mining.

Abstract: We infer moods and how they vary with time using a dataset from the social media application Twitter. We

used Python text mining techniques to gather all tweets originating from the Philippines within a span of 24

hours. From the dataset of around 130,000 tweets, we gathered the highest-frequency words and filtered out

neutral words to come up with words that imply mood levels. We then plotted the density of keyword usage

with respect to time, distinguishing between positive and negative moods. Our initial results of positive mood

and negative mood trends are consistent with published studies regarding microblogging mood scales. The

emergence of Big Data and the Internet of Things has greatly amplifed our ability not only to express ourselves

but to understand each other.

1 INTRODUCTION

Moods affect physical and emotional well-being,

creativity, decision-making, and immune response

(Ashby, et al, 2002). Studies have shown that mood

swings within the day heavily affect circadian

rhythms of core body temperature (Boivin et al.,

1997). In particular, positive moods have been found

to peak twice: at noon and at evening (Hasler et al.,

2008). Meanwhile, motivation is defined as a drive to

behave in certain ways that comes from internal and

external drivers and rewards such as mood (Deci and

Ryan, 1985). Motivation is therefore intertwined with

diurnal mood swings in correlation to human

productivity.

Data mining enables us to make real-time

measurements of global moods, such as social and

political issues (O’Connor et al., 2010). In particular,

researchers use the social networking service Twitter

as “a global thought-stream on every topic

imaginable” (Parr, 2009) with data gathering and

analytical tools available to the public. Mislove et al.

(2011) used Twitter comments to measure the “pulse

of the nation” noting various time zones and days of

the week. Meanwhile, Golder and Macy (2011)

examined Twitter mood trends across diverse cultures

accounting for global seasonal mood rhythms and

individual-level diurnal moods. Bollen, Mao and

Zeng (2011) used large-scale Twitter feeds to

measure collective mood and predict the stock

market. Self-reports of positive or negative moods

may be conveyed through microblogging sites such

as Facebook, Tumblr, or Twitter. Therefore

individual behavior in real time can be collectively

studied by examining semantic contents in such

sources.

The purpose of this study is to analyze circadian

moods of Filipinos using popular keywords with

mood implications using Python text mining. This

paper explored mood “pulses” confined within the

coordinates of the Philippines by examining the

density of popular keywords in tweets on an hourly

basis. The gathering of tweets spanned for one

weekday, or 24 hours.

Our study is limited only to publicly accessible

accounts, in accordance with data privacy laws. Also,

tweets are not representative of all age groups, as

these are limited to retrospective self-reports of the

younger generation (mostly millennials) and a low

percentage of older people (Newberry, 2018). Lastly,

this work will analyze word frequency only, looking

at the literal meaning of tweets; figures of speech such

as metaphors and sarcasm may be studied with more

advanced language processing tools.

2 METHODOLOGY

We used a Python text mining program (Figure 1) to

collect tweets in a streaming fashion via Twitter API.

364

Balajadia, R., Maglambayan, V. and Pulido, M.

Analysis of Filipino Mood Swings within a Day using Tweets.

DOI: 10.5220/0007749803640369

In Proceedings of the 4th International Conference on Internet of Things, Big Data and Security (IoTBDS 2019), pages 364-369

ISBN: 978-989-758-369-8

Figure 1: Flowchart of the program.

Figure 2: Sample tweet.

Analysis of Filipino Mood Swings within a Day using Tweets

365

We gathered tweets posted within a particular 24-

hour period, and only those originating from the

Republic of the Philippines. Aside from the message

itself, each tweet contains the handle or username of

the sender, the date and location from where it was

posted, the number of times it was retweeted and

favorited, and other useful information (Figure 2).

We obtained the frequency of each word, or the

number of times that a word appeared in the dataset.

For the purposes of this study which focuses on mood

swings, we manually filtered out fillers, conjunctions,

and other emotionally neutral words.

We manually separated the resulting high-

frequency words into two categories: those that

correspond to positive mood (PM) and those that

correspond to negative mood (NM). We plotted the

frequencies of these PM and NM words with respect

to time to look for possible trends corresponding to

diurnal mood swings.

3 RESULTS AND DISCUSSION

We gathered approximately 130,000 tweets within a

24-hour period from the Philippines. Figure 3 shows

the 30 words with the highest frequency. These

include words evoking PM or NM, but many of the

words are neutral or lack meaning.

Figure 3: High-frequency words (unfiltered).

Figure 4: High-frequency words (filtered).

IoTBDS 2019 - 4th International Conference on Internet of Things, Big Data and Security

366

After filtering out neutral words, the 30 resulting

high-frequency words became more evident in terms

of the personal moods that they convey (Figure 4).

We note that there were significantly more high-

frequency words that evoked NM rather than PM.

The words were taken literally, so curse words

like “tangina” were deemed negative while “lol

(laugh out loud)” was deemed positive. Although

“

☹

” is not exactly a word but an emoticon, we

decided to include it as it also expresses a specific

mood. There were still some neutral words such as

“okay” and “bukas (tomorrow or open)” which were

not considered as mood words. There were also words

that conveyed deep emotion such as “grabe (very)”,

but cannot automatically be categorized as positive or

negative.

Figure 5 shows the occurrence of the common PM

words within the observation day. The number of PM

tweets peaked during morning hours starting from 4

AM to 9 AM, where 9 AM had the highest peak. We

note that the circadian sleep component was found to

be elevated at waking and declines throughout the day

(Boivin et al., 1997).

Meanwhile, PM tweets peaked at the usual lunch

hour then declined during the afternoon (Figure 6).

PM tweets gradually rose to two of its highest peaks

during 7 PM and 9 PM, the common after-work

hours, and had the steepest dip during post-midnight

hours. In comparison, PM intuitively declined during

work hours in a study by Golder and Macy (2011).

Figure 5: Positive Mood (PM) Time Scale.

Figure 6: Negative Mood (NM) Time Scale.

Analysis of Filipino Mood Swings within a Day using Tweets

367

NM words were a constant plateau during

morning hours, gradually rising without relevant dips

during the afternoon, until it reached its peak during

after-work hours starting from 6 PM and peaking at

its highest at 8 PM. A continuous decline was

observed during sleeping hours, with a constant low

plateau during post-midnight hours, and once again

rising in the morning with a slight peak during 6 AM.

This is relatively consistent with existing studies with

their observation of the peaks and dips of NM (Golder

and Macy, 2011).

Figure 7 combines the PM and NM words used in

the previous two graphs. PM peaked during morning

hours while NM had a relatively low plateau, but both

consistently rose during after-work/after-school

hours. PM peaks and dips were more abundant than

the constantly rising and dipping scale of NM. Both

had its steepest dips during post-midnight/sleeping

hours, and both consistently rose as morning hours

approached. Despite the abundance of NM in

comparison to PM words, PM words were more

frequent during morning hours. However, NM had its

highest peak during after-work hours in comparison

to PM.

We can delve deeper into this data by obtaining

the high-frequency words that occur per hour, and

then determine if they indicate positive or negative

moods. Going further, we can determine the mood

evoked by each tweet, which will give a much more

accurate picture of mood and mood swings rather

than individual words. We may also use sentiment

analysis software to determine mood at a higher speed

(especially for large datasets), and to ensure that the

results are independent of possible bias from the

researchers.

We note that the data was gathered on a Tuesday,

which is commonly a work or school day. We can

extend the study to investigate trends for other days

of the week, or to consider user location,

demographic traits such as age or gender, social

network connections, and other related factors.

4 CONCLUSIONS

In conclusion, positive mood words exhibited erratic

peaks and dips during morning hours and gradually

declined during afternoon hours until it slowly rose to

achieve its highest peak during after-work/after-

school hours. Negative mood words consistently rose

from morning until its highest peak during after-work

hours without much erraticism. Despite negative

mood words outnumbering positive mood words,

positive mood words exhibited the highest frequency

during morning hours. Both moods exhibited steep

declines during post-midnight hours and consistently

rose as morning approaches. Such trends found for

tweets sent in the Philippines are consistent with

previous studies involving Twitter users in other parts

of the world.

We can expand this study by gathering data over

a longer time period, or determine if there are

regional differences in mood swings or at least in the

words used to express them. We can also use

sentiment analysis tools for more automated,

objective measurement of positive and negative

moods. The emergence of Big Data and the Internet

of Things has greatly enhanced not only our ability

to express ourselves but also the ability to

understand each other.

Figure 7: Comparison of PM and NM Time Scales.

IoTBDS 2019 - 4th International Conference on Internet of Things, Big Data and Security

368

ACKNOWLEDGEMENTS

We thank the Mapúa University Yuchengco

Innovation Center for the resources in preparing this

manuscript, and our colleagues and loved ones for

their support. We also thank the organizers of the

IoTBDS 2019 Conference for accepting this work and

for the financial support.

REFERENCES

Ashby, F. G., Valentin, V. V. & Turken, A. U., 2002.

Emotional cognition: From brain to behaviour.

Boivin, D. B., et al., 1997. Complex interaction of the sleep-

wake cycle and circadian phase modulates mood in

healthy subjects. Archives of general psychiatry, 54(2),

pp.145-152.

Bollen, J., Mao, H. & Zeng, X., 2011. Twitter mood

predicts the stock market. Journal of computational

science, 2(1), pp.1-8.

Deci, E., & Ryan, R. M., 1985. Intrinsic motivation and

self-determination in human behavior. Springer

Science & Business Media.

Golder, S. A. & Macy, M. W., 2011. Diurnal and seasonal

mood vary with work, sleep, and daylength across

diverse cultures. Science, 333(6051), pp.1878-1881.

Hasler, B. P., et al., 2008. Preliminary evidence of diurnal

rhythms in everyday behaviors associated with positive

affect. Journal of Research in Personality, 42(6),

pp.1537-1546.

Mislove, A., et al., 2011. Understanding the Demographics

of Twitter Users. ICWSM, 11(5th), 25.

Newberry, C., 2018. 28 Twitter Statistics All Marketers

Need to Know in 2018. [Online] Available from:

https://blog.hootsuite.com/twitter-statistics/ [Accessed

30 December 2018].

O'Connor, B., et al., 2010. From tweets to polls: Linking

text sentiment to public opinion time series. Icwsm,

11(122-129), 1-2.

Parr, B., 2009. 5 Terrific Twitter Research Tools. [Online]

Available from: https://mashable.com/2009/05/03/twit

ter-research-tools/ [Accessed 03 January 2019].

Analysis of Filipino Mood Swings within a Day using Tweets

369