DISC 325

Business Data Management

Project Report

Group 12
Muhammad Ahmad Najam 22110252
Ali Shahid 22110194
Ahmad Muneeb 22110184
Ahmed Zeeshan 22110340
 Table of Contents
Section Page #

1. Background 1

2. Methodology 1

2.1 Data Extraction 1

2.2 Data Cleaning 2
2.3 Data Visualization 3

3. Analysis 3

3.1 Correlation Matrix 3

3.2 Tweet Analysis 4
3.3 Heat Map 7
3.4 Sentimental Analysis 11
3.5 General Characteristics of the Data 12
3.6 Bigrams and Trigrams 14
A) Positive Sentiments 15
B) Negative Sentiments 16

3.7 Dubai Corundum and the Word ‘Young’ 18

4. Work Cited 20
 Group 12

1: Background:
This twitter data revolves around the time period of October 2019 from United Kingdom.
During this time two major political events were aligned to be held in the near future.

1. Britain Elections: The Britain elections were scheduled to happen in December

2019 where two of the major parties, Conservatives and the Labour, were competing
in the elections. Both of these parties had different agendas on their manifestos out of
which one was the closure of the Brexit deal. The Conservatives wanted to close the
deal by the end of January 2020 while the Labour wanted to hold a referendum in six
months’ time and act according to the decision of the people.

2. The Brexit Deal: The Conservative was adamant on closing the Brexit deal
between UK and EU at the earliest. It was decided that UK will be leaving the EU
under certain terms related to trade, immigrations and other economic activity.
Leaving EU meant that UK can no longer trade freely with EU and its citizen can no
longer enjoy the autonomy of working in EU without any significant restrictions. The
Conservative party and EU were trying to negotiate on new rules of living, working,
and trading between the two, which attracted a lot of debate on social media.

2: Methodology:
The data file that we received had raw unorganized data in which all the attributes resided in
a single row. The file size was 461 MBs when we received it, which we reduced to 21 MBs.
This was done in three major steps.

2.1 Data Extraction

The original file had around 56 columns in which different attributes were stored.
During preliminary examination we decided to extract only 48 columns that we thought
were useful for our analysis. For example, from the original data we did not extract the
URL column. We decided this on a qualitative basis while doing the initial examination
of the data file.

Initial examination also allowed us to realize that the name of the columns was stored in
each cell before the data, separated by a semi-colon (:). Firstly, we made an array of the
name of the 48 columns that we decided to extract from the original data file. Next, we

1
 Group 12

used three main functions; Strip, Replace and Count to read the data line by line and
extract the data after the initial semi-colon and store it in a new file to create rows
corresponding to each tweet. The data was arranged in a table form with different
column headings. After this was done, then we went on our Data Cleaning Phase.

2.2 Data Cleaning:

The Data in the CSV file was cleaned and sorted as follows:
• We removed all those rows in which the field of location was empty. Since location
was one of the main attributes, we used for data visualization, tweets with no location
value were removed. Apart from this we also removed repeated rows from our data
set to make it less redundant, reducing the row count to 15,000 rows from 72,000
rows originally.

• We reduced the number of columns to 23, keeping the ones we thought contained
relevant data for our analysis. The columns which had string values were removed
since they were of no use in establishing a relationship. Apart from this the columns
which had null values were also removed using a python’s code.

• A new CSV file was created with clean and formatted data which we used further for
data analysis.

• The new CSV file contained the following columns:

➢ created_at: Shows the date and time at which the tweet was made.
➢ Text: contains the text of the tweet made.
➢ Source: contains the web location of the tweet.
➢ in_reply_to_screen_name: shows if the tweet is a reply to someone.
➢ User-id: the user ID of a person who posted this tweet.
➢ Name: Name of a person who made the tweet
➢ Screen_name: Screen name is the username of the twitter account, Which the
users choose to identify themselves on the website.
➢ Location: he location at which the tweet is made
➢ Verified: shows if the account is verified.
➢ followers_count: Number of followers
➢ friends_count: number of people following.
➢ listed_count: The number of public lists that this user is a member of

2
 Group 12

➢ favourites_count: Number of tweets liked by this user.

➢ statuses_count: Number of tweets made from that account.
➢ Geo-Enabled: This field must be true for the current user to attach geographic
data while posting statuses.
➢ quote_count: This tells us how many times other users have quoted this tweet.
➢ reply_count: The number of replies on the tweet.
➢ retweet_count: The number of times the tweet has been retweeted.
➢ favorite_count: Number of likes received on the tweet.

2.3 Data Visualization:

After extracting and cleaning the data, we started conducting our analysis. For that we relied
mainly on visualizations as they are easy to interpret and analyze. Furthermore. It also
allowed us to capture any patterns in the data like correlations that might give us some key
insights about the data. All these are discussed in the next section.

3: Analysis:

3.1 Correlation Matrix:

The correlation matrix represents

the relationship between different
attributes in the data. From this
correlation matrix we can observe
a few relations between different
attributes in our data set.

Figure 1:Correlation Matrix

3
 Group 12

• Retweet Count and Favorite Count: A strong relation is observed between these two
variables. This is mainly because when a person likes a tweet (Favorite), there is a
high chance that person will also retweet it on their profiles for others to see.
• Reply Count and Quote Count: A relatively less strong relation can be seen between
these variables. A reason for this is that when a person replies to a tweet, twitter
automatically quotes the person who is being replied to hence there is relation
between these two numbers.
• Retweet Count and Quote Count: Another slightly strong relation was observed here
because when a person retweets, the author of the tweet is quoted along in that
retweet. Therefore, there is a positive relation between these two numbers.
• Favorite Count and Quote Count: A relationship can be observed here as when a
person likes a tweet there is a chance that they will comment on it too. While
commenting the author of the tweet is automatically quoted.
• Listed Count and Followers Count: There is a very strong relation between these
variables mainly because when a person is parts of any lists there is a high chance that
other people of that list will follow this person, hence the strong relation.
• Follower Count and Status Count: A relation here can be explained in a way that
when a person begins to tweet actively on twitter there is a chance that his followers
will increases.

3.2 Tweet Analysis:

The graph on the right shows

the top 12 users with most
followers. Among these, we
noted that all off the accounts
are verified except for ‘EL
MUNDO’. Apart from this,
“The Economist” has the
greatest number of followers
followed by ‘CGTN’. It is
worthwhile to note that only
Figure 2: Top 12 Users
1.9% of the accounts in our data
set were verified. The pie chart below depicts this clearly.

4
 Group 12

Figure 3: Verified vs Non-Verified Users (Left)

Default Profile Users (Right)

Apart from this there were 56% of the default profiles in our data. This means that these 56%
of the accounts have not edited anything on their accounts such as the background or the
profile picture etc. A very high percentage of default profiles and a very low number of
verified accounts indicate that there were a lot of bot accounts in our data set. This leaves a
lot of room for the manipulation of the masses since bot accounts can be used to tweet a
certain topic and make it trending on twitter.

Figure 4: Most Retweets by Users

5
 Group 12

The above displayed bar chart shows the users who have the highest number of retweets.
Leading the race is the user ‘_entropea’ who had 85,784 retweets on the tweet he made about
the extension of Brexit deal. It is surprising the tweet was not made from one of the hotspots
on heat map that were situated in the UK, rather, it had been made from the French capital of
Paris. It is important to note that as established earlier, likes and retweets have a strong
correlation. This relation is testified by the results that are presented here. We can see this
since tweets with most retweets are also the ones which got vast number of likes on them.

Figure 5: Tweet with the most Likes

The above figure clearly shows the most liked tweets in our data set. The strong correlation
between most liked tweets and most retweeted tweets can be observed. Tweets that have
highest retweets are the ones which also have one of the highest likes on twitter. This also
makes sense as many people tend to give out more attention to tweets that the rest of the
public is more attentive on. Thus, we can somewhat conclude that tweets particularly on this
topic are liked by other users who make them more visible for other users by retweeting.
This creates a cycle of sorts whereby specific tweets gain both likes and retweets
simultaneously.

6
 Group 12

3.3 Heat Map:

Figure 6: Heat Map

With the help of python, we cleaned the location data for null/bogus entries and used
packages: GEOPY and NORMADIN to develop two new datasets containing the locations
and their respective coordinates. These coordinates were plotted onto the world map with a
centric focus on the United Kingdom (UK). The heatmap formed shows us the distribution of
the tweets across the world for our dataset. We gained many insights from this.

Figure 7: Most Number of Tweets by City

7
 Group 12

This chart above shows the top 20 cities where the tweets were made using only conventional
cleaning methods. It is worth noting here that the graph shows England, United Kingdom and
UK as cities while they are in reality not solely due to the limitation of user self-reported
locations that were in our dataset and as lot of fields in the location attributes did not have a
precise location. Instead, they contained the country in which the tweet was made. This
meant that this bar graph with topmost active locations is only marginal in usefulness hence
we dug deeper to gain useful insights which we eventually did using the aforementioned
packages. The insights are as follows:

• Europe-Centric: This is a primarily Europe-centric issue with little interest outside

of it apart from maybe concerns regarding trade implications. However, the term
Europe-centric is applied rather loosely here as we can see in the below mentioned
figure of our heat map that conversion regarding Brexit was largely more active in
Western Europe than its Eastern counterpart. This can be attributed to more authority
that the Western powers possess in the halls of power of the EU as compared to other
smaller Eastern European nations.
• The Big Three: Most tweets come from UK, France, and Germany which are the
big three powers of Europe. As illustrated by news reports at that time, the most
active in this debate were these three nations with their leaders e.g., Angela Merkel,
Emmanuel Macron, and Boris Johnson, each at odds with one another. This is
supported by the distribution shown in our heat map and shows how polarized the
European powers had become over the issue. Indeed the influence of these political
stalwarts meant that each country tended to engage in the discussion with equal zest.

Figure 8: The major hotspots in Europe.

8
 Group 12

• The US: The US is not as involved as one might initially expect with only major
cities shown as hotspots. This makes sense as the issue did not affect the US beyond
minor trade implications. Another interesting angle to note from this is that the US is
still tied to Europe and its powers through the provision of NATO. This means that
the USA’s diplomatic ties and the ability to muster an armed force from the alliance
remain unaffected. Thus, it doesn’t make sense for the US to politically involve itself
in the issue apart from seeking minor trade assurances for its own citizens. Another
reason is that the US even apart from NATO has kept individual diplomatic ties with
London, Paris and Berlin. Hence, forwarding the interests of one group out of these
could have meant serious dissent from the other two allies.

Figure 9: US Heat Map

• Where Does Dubai Factor in: Interestingly, we see that as revealed by our word
cloud (it is mentioned in the following section), Dubai is also a hotspot for the tweets
which means that citizens in the UAE are perhaps more concerned with the issue than
their other middle eastern and south Asian counterparts. While this will again be
discussed with more detail later on in the report, however, it is worth noting here that
this minor hotspot alludes to a much bigger phenomena. As the raw dataset had been
primarily composed of tweets in English and even later cleanings that were conducted
using NLTK and Corpus packages meant that there exists a possibility of there being
a much more undermined hotspot in Dubai contrary to the case in reality. A holistic
view of news from that timeframe meant that our earlier prediction of there being
discourse in Dubai regarding Brexit more than its Middle Eastern and Asian
counterparts proves to be true.

9
 Group 12

Figure 10: Dubai Heat Map

• South Asia: Apart from the earlier mentioned hotspots, little discourse is there in
South and South-east Asia. Where little there is, it is restricted to mostly major cities
and corporate hubs. However, even in this, we see that ports occupy more hotspots
which again points to the earlier made hypothesis that this is primarily a Europe-
centric issue and the rest of the world has little interest apart from concerns about
trade implications.

Figure 11: Heat Map of South Asia

10
 Group 12

3.4 Sentimental Analysis:

To have some deeper insights of the

data, we had used the NLTK package’s
Vader library to use the “Sentiment
Intensity Analyzer” function. The reason
why we chose this library rather than the
traditional NLTK one is due to the fact
that this provides us with what is a pre-
trained sentiment analyzer. While on
may argue with the relative effectiveness
of this when compared to a self-
Figure 12: Sentiment Distribution of the data.
developed sentiment analyzer, it should be
noted that the project’s scope of work and time constraints did not allow for such a time-
consuming measure. We were also limited by the resources we had as we clearly lacked a
classification dataset to use for development. Thus, we went with the aforementioned Vader
package for our analysis. Initially when we ran the analyzer, our data was extremely skewed
towards Neutral side. This we found could be attributed with the methodology of how the
Vader package worked. As our data contained a lot of duplicates and retweets, phrases
which otherwise would have induced a negative or positive score came out to be neutral. To
combat such a case, we further cleaned our data while ensuring that this did not bias our data
in any way. We can confidently confirm this as this was done with the rules regarding
random sampling in mind. This measure also had the added bonus of reducing the influence
of bot accounts which tend to tweet the same text multiple of times. Thus, we treated the
skewness by retaining only the original tweets. Once we did that, we got a balanced graph
with majority of the tweets being neutral having a share of 41.5% whereas there was almost
equal division between positive and negative sentiment. The graph is shown above. This
result was in coherence to the results of the Brexit referendum that was held two years
earlier by the government and also with the state of the parliament which was by all extents
extremely indecisive and termed as a “Hung Parliament”. Using these results, we did further
analysis by generating word clouds, bigrams, and trigrams. These are given below:

11
 Group 12

3.5 General Characteristics of our Data:

We drew the Word Cloud for our data that enabled us to get an idea of the topic of the
conversation that were discussed in the tweets. The bigger the size of the word, the more
number of times it came up in the data. Thus, we were able to pick out words whose usage
was frequent and very relevant to our analysis. Some of these were expected e.g., “Brexit”,
“Vote”, “party”, etc. however, some words came to us as a surprise. These will be brought
up in proceeding sections.

Furthermore, the prominence of these words is Most Used Words are:

coherent other descriptors we had run after 0

brexit 692
successfully tokenizing the data and using the vote 199
elect 159
“COUNTVECTORIZER” package. This frequency uk 126
parti 123
of words can be seen from the data that is shown in tori 117
Figure. 11. Again to reemphasis, many of these
Figure 13: Number of Words
words were anticipated as the data revolves around
the Brexit decision and the elections that were
going to be held in the United Kingdom. The most The Average length of Tweet text is:

text_len
interesting insight that we got from this analysis sentiment
negative 105.63
was the appearance of the word “Dubai” and neutral 91.39
positive 103.84
“Young”. This will be discussed in section 3.7 of
the report. The Average words in Tweet text is:

text_word_count
To further add depth to our analysis, we also sentiment
negative 16.59
neutral 11.48
categorized using a combination of count and group positive 16.44

by functions to get the most used words according

to which sentiment they belong to and to see what Figure 14: Avg Length by sentiment

intricacies of language entice different reactions.

We then also analyzed the average length of the tweet according to the sentiment. Figure 12
allows us to imply that people with either positive or negative sentiments, tend to be more
vocal than neutral ones owing to the fact that they tend to express themselves more hence
the significantly bigger word counts and text lengths.

12
 Group 12

If we analyze the following two figures each belonging to a polarized sentiment, we see an
apparently confusing dilemma. The words used are more or less the same as can be seen
clearly that words like “Vote” and “Elect” are listed both as positive and negative. This
meant that there was a need to perform greater analysis and that utilizing only word clouds
would again limit us to only a marginally useful analysis as was the similar case in the
location distribution. Thus, this led us to our next step of the analysis namely the
employment of n-grams: Bigrams and Trigrams.

Most Used Positive Sentiment Words are: Most Used Negative Sentiment Words are:

0 0
brexit 205 220
parti 106 vote 55
vote 88 elect 52
tori 60 uk 38
elect 55 deal 35
want 50 tori 34

Figure 15: Word Cloud

13
 Group 12

3.6 Bigrams and Trigrams:

Bigrams (or Combinations of two words) are:

[('u2026 https', 228), ('brexit party', 60), ('brexit deal', 51), ('general election',
50), ('boris johnson', 49), ('deal means', 31), ('means u2018', 31), ('u2018 u00a370bn',
31), ('u00a370bn hit', 31), ('uk 2029', 30), ('brexit https', 29), ('stop brexit', 22),
('hit uk', 22), ('brexit election', 21), ('2029 https', 21), ('brexit u2026', 20), ('don
u2019t', 20), ('lib dem', 18), ('u03c4 u03bf', 18), ('vote brexit', 16)]

Trigrams (or Combinations of three words) are:

[('deal means u2018', 31), ('means u2018 u00a370bn', 31), ('u2018 u00a370bn hit', 31),
('brexit deal means', 30), ('u00a370bn hit uk', 22), ('hit uk 2029', 21), ('uk 2029
https', 21), ('boris johnson u2019s', 11), ('bbc news brexit', 11), ('vote brexit party',
10), ('johnson u2019s brexit', 10), ('news brexit deal', 10), ('u00a370bn hit year', 9),
('hit year uk', 9), ('year uk 2029', 9), ('ud83c uddec ud83c', 9), ('johnson brexit
deal', 9), ('brexit u2026 https', 9), ('ud83d ude02 ud83d', 9), ('ude02 ud83d ude02', 9)]

We made Bigrams and Trigrams for each type of sentiments. The above figures represent
only two of the eight such exhibits that we had designed. For the ones shown, these belong
to the overall combination of words/phrases that were used and not to a particular sentiment.
That we shall explore in the last sections of this report. However, here we can see the utility
that the usage of bigrams and trigrams bring to our analysis as they detail how the firsts were
used apart from what words were used. They also illustrate the number of times a particular
phrase or combination of words has come up which further add richness to our analysis. For
example, we can see that “Brexit” even if it may represent the exit of the UK from the EU,
means something entirely else when used as “Brexit Party” as now it refers to the
conservative party which had been the biggest advocates of this deal. Another such example
is the word “hard” which might appear meaningless in the wordcloud but when coupled with
the other words such as “Hard Brexit” instantly make more sense as this referred to the state
where the UK would leave the EU without any working deal. Thus, this helps us go the extra
mile in our analysis which a typical word cloud cannot hope to match.

14
 Group 12

A) Positive Sentiments:

In this section, we will elaborate on the intricacies revealed by our analysis of only
positive sentiments which includes visualizations such as word cloud, bigrams and
trigrams. Below is drawn the word cloud for only positive sentiments.

Figure 16: Positive Sentiment Word Cloud

Here we see that words such as “Brexit”, “Vote”, “Party”, etc. are all shown as those
which induce a positive sentiment. However, as mentioned in detail in the preceding
sections, this only has marginal usefulness. These words essentially just appear a lot in
our dataset and thus the above word cloud can be a bit misleading for the uninitiated.
Thus, to tackle this problem, we again turn to the usage of our other visualizations that
are bigrams and trigrams. These are shown below:

Bigrams (or Combinations of two words) for positive sentiments are:

[('u2026 https', 95), ('brexit party', 52), ('general election', 20), ('boris johnson',
15), ('vote brexit', 11), ('u26a0 ufe0f', 10), ('stop brexit', 10), ('don u2019t', 10),
('brexit u2026', 9), ('tory party', 8), ('ufe0f u26a0', 7), ('th u2026', 7), ('corbyn
labour', 6), ('https u2026', 5), ('tory brexit', 5), ('labour party', 5), ('brexit
election', 5), ('nigel farage', 4), ('u2019t want', 4), ('brexit deal', 4)]

Trigrams (or Combinations of three words) for postive sentiments are:

[('vote brexit party', 9), ('u26a0 ufe0f u26a0', 7), ('ufe0f u26a0 ufe0f', 7), ('don
u2019t brexit', 4), ('brexit u2026 https', 4), ('corbyn labour party', 4), ('u201cthe
tories don', 3), ('tories don u2019t', 3), ('u2019t brexit nand', 3), ('brexit nand
tory', 3), ('nand tory party', 3), ('tory party far', 3), ('party far brought', 3), ('far
brought treaty', 3), ('brought treaty think', 3), ('treaty think good', 3), ('think good
u201d', 3), ('good u201d u2026', 3), ('deal means u2018', 3), ('means u2018 u00a370bn',
3)]

15
 Group 12

This provides a much more insightful perspective into the topic as this shows the
combinations and phrases that were used and successfully induced positive sentiments
about Boris Johnson’s government and the whole Brexit affair. The most notable of
these are described below:

• [Brexit Party]: Used a lot, this represents the conservative party which at the time
was also the ruling party and the main advocates of the whole Brexit idea. This
shows that the people were generally more positively inclined towards mentioning
the conservative party in a good light.
• [Boris Johnson]: The leader of the conservative party and the presiding PM. Boris
Johnson was surprisingly also described roughly in the same manner e.g.,
represented both in a positive light and in a negative one. This shows the polarizing
effect of the personality as even though people might support the conservatives, they
however, were still apprehensive of the often-eccentric Boris.
• [Corbyn Labour Party]: Jeremy Corbyn is the leader of the main opposition
party to the conservatives e.g., the labour party. His presence in the positive
sentiments of people can actually be explained by the recent events preceding this
dataset. After a long-drawn-out debate with little to no real results in what Boris
Johnson had described as the “hung parliament”, in almost a surprise move, Corbyn
agreed to the notion of December elections. This move signaled the end of long-
drawn-out discussions and the possibility of real progress on the issue to finally take
place. Thus, even though political allegiances might have divided the people, this
move by Jeremy Corbyn was largely welcomed. More than anything, it indicated
how exhausted were the people of UK from being stuck in a limbo like state.

B) Negative Sentiments:

In this section, we will elaborate on the intricacies revealed by our analysis of only
negative sentiments which includes visualizations such as word cloud, bigrams and
trigrams. Below is drawn the word cloud for only negative sentiments.

16
 Group 12

Figure 17: Negative Sentiment Word Cloud

Here, we can see the exact same words as before occupying the most prominent of
positions. Thus, this word cloud like the preceding one for positive sentiments is also
only marginally useful. That is why we shall again turn towards the usage of both
bigrams and trigrams to reveal what combinations of words and what phrases were
inducing negative sentiments.

Bigrams (or Combinations of two words) of negative sentiments are:

[('u2026 https', 68), ('boris johnson', 14), ('general election', 14), ('brexit deal',
13), ('stop brexit', 12), ('lib dem', 11), ('brexit https', 10), ('brexit news', 9),
('brexit party', 8), ('deal leave', 8), ('brexit u2026', 7), ('deal brexit', 7), ('leave
uk', 7), ('uk u00a370bn', 7), ('u00a370bn worse', 7), ('don u2019t', 6), ('hard brexit',
6), ('news lib', 6), ('dem mp', 6), ('mp humiliated', 6)]

Trigrams (or Combinations of three words) of negative sentiments are:

[('deal leave uk', 7), ('leave uk u00a370bn', 7), ('uk u00a370bn worse', 7), ('brexit
news lib', 6), ('news lib dem', 6), ('lib dem mp', 6), ('dem mp humiliated', 6), ('mp
humiliated twitter', 6), ('humiliated twitter brexit', 6), ('twitter brexit claim', 6),
('brexit claim mocked', 6), ('claim mocked leavers', 6), ('mocked leavers uk', 6),
('brexit deal leave', 6), ('u00a370bn worse year', 6), ('boris johnson u2019s', 5),
('ud83d udc46 ud83d', 5), ('udc46 ud83d udc46', 5), ('leavers uk news', 4), ('pm brexit
deal', 4)]

17
 Group 12

This, as it also did in the preceding section, gives a much more insightful look into negative
sentiments of people that tweeted in our dataset. The most notable of these are described
below:

• [Boris Johnson]: The coming up of his name suggests our earlier observation of
him being a very polarizing personality for people on both sides of the argument holds
true.

• [Brexit Deal]: Many people in the UK as well as those outside it were either not
satisfied by the provisions of the deal or thought of them as excessive. Thus, the fine
details of the deal did not appeal to many people either in the UK or in the EU.

• [Dem MP Humiliated]: Apart from the two major parties e.g., the conservatives
and the labour party, this third party of the Liberal Democrats had its own opinions.
While Jeremy Corbyn had decided to go with the idea of early elections, the liberal
democrats where not that keen and cited a deep mistrust in Boris Johnson. Thus, this
party is portrayed in a negative light. Furthermore, with Corbyn shifting his stance to
align with Boris, meant that the opinion of the Liberal Democrats became
unnecessary as they did not have considerable power in the electorates. This explains
why they were “humiliated”.

• [Hard Brexit]: This was found to induce negative sentiments as this signified
circumstance where Britain would leave the EU regardless of whether a deal has been
reached or not. This was vehemently opposed by all people on both sides of the
argument.

3.7 The Dubai Corundum and the Word ‘Young’:

Dubai is very close to Europe when in regards to trade relations and hence has international
ties with many of the various states that together form the European Union. The whole
Brexit situation would directly impact the Dubai’s economy as they had plans for future
expansion into the UK and from there onwards into more mainland Europe. This is testified
by entries in our data file which show that Dubai was looking for an expansion plan in the
UK and that this idea was shelved indefinitely until the uncertainty around the Brexit issue
had not dissipated. This also explains why Dubai came up so many times in our word cloud
and bigrams/trigrams.

18
 Group 12

The word young had earlier also been identified as one of the more prominent words in the
word cloud. This came up in the arguments that people were having that “a third of young
people are not registered to vote”. One explanation of this is that Brexit would have the most
impact on the younger generation which was still coming of age. The individuals in this age
bracket were typically 16-21 years old and were included in the referendum but strangely
not in the general election. Given that this is the generation that is the future of the country,
and the fact that most of them are not even registered to vote on an issue that will affect
them directly was a point of contention for not only common folk but with members of the
parliament as well. This insight is a particularly big eye opener for the government that
inconsistencies in constituting voter counts for both elections and referendums can be very
harmful as they can bias the results and create dissent among the excluded masses.

These intricate insights more than anything also shed light on how complex of an issue
Brexit truly is and the need for a collective united response from the stakeholders unlike that
was seen in the 2019 and 2020 in Britain. In hindsight, they must think about Brexit wisely
as the decision would impact both the future generation and the economy of the country
directly.

19
 Group 12

Work Cited

• Luvsandorj, Zolzaya. “Simple Wordcloud in Python.” Medium, Towards Data

Science, 8 Dec. 2020, [Link]/simple-wordcloud-in-python-
2ae54a9f58e5.
• Yener, Yalin. “Step by Step: Twitter Sentiment Analysis in Python.” Medium,
Towards Data Science, 7 Nov. 2020, [Link]/step-by-step-twitter-
sentiment-analysis-in-python-d6f650ade58d.
• “What's in the EU-UK Brexit Deal?” Council on Foreign Relations, Council on
Foreign Relations, [Link]/in-brief/whats-eu-uk-brexit-deal.
• Lane, Chloe. “The UK 2019 General Election Party Manifestos: Labour vs
Conservatives.” Top Universities, [Link], 14 Apr. 2021,
[Link]/where-to-study/europe/united-kingdom/uk-2019-general-
election-party-manifestos-labour-vs-conservatives.

20