Available via license: CC BY
Content may be subject to copyright.
The Impact of Incumbent/Opposition Status and Ideological Similitude on
Emotions in Political Manifestos
Takumi Nishi1
1Waseda University, Japan
Corresponding author: Takumi Nishi , t-nishi@fuji.waseda.jp/taknishi95@gmail.com
Abstract
This study analyzes emotion-associated language in the United Kingdom’s Conservative and Labour
Parties’ general election manifestos from 2001 to 2019. While previous research has shown a cor-
relation between ideology and positions in policy, there are still conflicting results in matters of the
sentiments in such manifestos. Using new data, we present how a manifestos’ valence level can be
swayed by a party’s status in government, with incumbent parties presenting a higher frequency in posi-
tive emotion-associated words while negative emotion-associated words are more prevalent in opposition
parties. We also demonstrate that parties with ideological similitude use positive language prominently,
further adding to the literature on the relationship between sentiments and party status.
Keywords
sentiment analysis; political manifestos; NRC Lexicons, VADER Sentiment Analysis, emotion-laden
words
I INTRODUCTION
In an age where social media and other digital communication tools are key mediums for mass
communications, various political parties have dedicated their effort to establishing their so-
cial media strategy and capitalizing on its strengths [Yildirim,2020]. And consequently, there
has been a growing perception within democratic societies viewing political manifestos as an
obsolete relic [Bogdanor,2015]. While it can be suggested that the impact of manifestos has
relatively waned over the decades, many scholars continue to emphasize their importance in
addressing various parties’ campaign strategies [Eder et al.,2017]. In particular, manifestos en-
able parties to differentiate their policy from their competitors while streamlining the campaign
strategy for its politicians. Additionally, recent studies on manifestos indicate the importance of
emotive language in the text as a crucial asset in persuading voters [Eder et al.,2017,Koljonen
et al.,2022]. In effect, we offer that political manifestos can provide crucial data to uncover a
political party’s strategy in deploying emotions, especially during elections.
While the recent interest in social media’s role in politics has produced many studies in sen-
timent analysis of social media posts such as tweets, there is little research on the same topic
using manifesto texts [Belcastro et al.,2020,Koljonen et al.,2022]. Using 12 general election
manifestos from the United Kingdom’s two establishment parties, the Conservative Party and
the Labour Party, published between 2001 and 2019 as data, we will utilize a lexicon-based
emotional analysis approach to investigate how a party’s position in government (incumbent
Journal of Data Mining and Digital Humanities
ISSN 2416-5999, an open-access journal
1http://jdmdh.episciences.org
arXiv:2305.08383v2 [cs.CL] 21 Sep 2023
or opposition), alongside, ideological similarities affect the use and frequency of emotive lan-
guages in manifestos over time.
II PREVIOUS RESEARCH AND HYPOTHESES
At the current level, previous manifesto research related to sentiment/emotion analysis revolves
around various European parities’ campaign materials [Crabtree et al.,2018,Jentsch et al.,2021,
Koljonen et al.,2022], of which many show evidence that the party’s status in government and
ideological positioning have a standing influence on these issues. Based on these findings, the
study establishes two hypotheses (H) as follows:
Hypothesis 1 (H1): The frequency of positive emotions will be highest in the incumbent
party’s manifesto than the opposition, while the frequency of negative emotions will be high-
est in the opposition party’s manifesto than the incumbent
A study conducted in campaign materials (including manifestos) across European elections in-
dicates a correlation between the usage of emotions and the party’s current status in government
[Crabtree et al.,2018]. To be exact, the party or parties who are part of the incumbent govern-
ment increase the frequency of positive emotions across different campaign materials than the
opposition. Regarding coalition governments, the party that controls the Prime Minister’s posi-
tion uses positive emotions more than its coalition partners [Crabtree et al.,2018]. The reason
incumbent parties, especially those that control the PM office, increasingly assert more positive
language. This can be related to the voters directly assigning them greater responsibility as the
position of Prime Minister is involved in the planning and execution of public policy, which
can then attract mass public attention and even more scrutiny from the general public [Crabtree
et al.,2018]. This incentivizes the incumbent parties to improve the voter’s perception of their
government and its past decisions by positively portraying themselves (especially in times of
national crisis or civil unrest).
However, in Finnish party programs, Koljonen et al. [2022] did not find differences between
the incumbent and opposition parties using emotion-associated words. Instead, they found the
differences statistically insignificant except for the populist parties who used significantly more
intense words to describe their party goals.
In terms of the opposition, due to its main role as a critic and a rival to the incumbents, it has
been shown to exploit negative sentiments and language within its campaign materials as a strat-
egy to undermine the incumbents’ image [Louwerse et al.,2021]. Based on these dynamics of
a party’s position in government and its impact on positive and negative language, the research
formulates H1.
Hypothesis 2 (H2): Manifestos of ideologically similar parties will use positive language
more frequently than negative language
Kosmidis et al. [2018] found that in an election between parties that are similar in ideology and
policy content (i.e., established centrist parties), they would compete using ”emotional appeals.”
Specifically, these parties would primarily adopt positively associated words to captivate the
same voter base [Kosmidis et al.,2018].
While the British Conservative Party and Labour Party have been recognized as ideological
opposites from the 1900s to 1970s, the increased approval of policies under Thatcherism and
Neo-liberalism starting from the 1980s in British politics have led to their gradual convergence
in ideology and policy [English et al.,2016]. This convergence has continued into the 21st
Journal of Data Mining and Digital Humanities
ISSN 2416-5999, an open-access journal
2http://jdmdh.episciences.org
century, with research finding that Labour Party and Conservative Party politicians continued
to share neo-liberal values even after the Great Financial Crisis and succeeding economic crisis
between 2008 and 2014 [English et al.,2016]. Based on this, the research establishes H2 in
which ideologically similar parties (the Conservative Party and the Labour Party) will have a
higher frequency of positive language than negative language in their manifestos.
Summary: Both H1 and H2 aim to investigate how a party’s government position (incumbent
or opposition) and ideological positioning (similarities) will impact the type of emotions and
their relative frequency in manifesto texts.
III DATA AND METHODS
In the UK, manifestos are only released in preparation for the nation’s general elections for
the House of Commons, where the results of seats of MPs will determine which party (or par-
ties) forms a government [Wikipedia,2022a]. Therefore, the data includes 12 general election
manifestos, six each for the Conservative Party and Labour Party, published over six general
elections (2001, 2005, 2010, 2015, 2017, and 2019). All manifestos have been accessed from
two Wikipedia pages, each dedicated to both parties respectively, that present lists of available
manifestos accessible by PDF files or redirected to websites with the manifesto texts (which
were later converted and downloaded as PDF files) [Wikipedia,2022a,b].
The decision to retrieve manifestos from 2001 to the most recent election of 2019 was to provide
enough data to observe any salient differences in the composition of emotive languages and
sentiments when both parties either switched as the incumbent or opposition. From 2001 to
2010, the UK was led by a Labour government, with the Conservative Party as the primary
opposition [Wikipedia,2022b]. Their positions switched with the outcome of the 2010 election,
where the Conservative Party formed a government still in power as of 2023 [Wikipedia,2022b].
Therefore, 2001 to 2019 offers a sufficient timeline to pursue the research objective effectively.
For the text extraction from PDF files, a PDF text extractor called ”pdfminer.six” was used
for this research. Unlike other PDF extractors such as PyPDF2, pdfminer can compensate for
two-column documents with relative success. After the extraction was successful, the NLTK
sentence tokenizer [Bird et al.,2009] was used to separate the large string of text into a list
of sentences, which were then processed, removing newlines and URL links and correcting
Unicode formats.
3.1 Method 1: VADER Sentiment Analysis
This study used a popular lexicon-based sentiment analysis tool ”VADER” by Hutto and Gilbert
[2014] to investigate each manifesto’s general composition of positive, negative, and neutral
sentences within a manifesto text.
VADER (Valence Aware Dictionary for Sentiment Reasoning) uses a lexicon-based approach to
analyze each lexicon within texts and return an aggregated sentiment score of the text ranging
from -1 to 1 (most negative and most positive respectively) [Calderon,2018]. To do so, VADER
utilizes a dictionary that assigns sentiment scores to different lexical features and emoticons
(i.e. emojis and colloquialisms). Moreover, VADER accounts for other text features, including
capitalization, punctuation, and negation, making it a popular lexicon-based sentiment analysis
tool, especially for social media texts [Calderon,2018].
The manifestos were left as is (no stopword removal, lemmatization, etc., were implemented),
and the texts (manifesto sentences) were processed by the VADER ’Sentiment Intensity Ana-
Journal of Data Mining and Digital Humanities
ISSN 2416-5999, an open-access journal
3http://jdmdh.episciences.org
lyzer’ object to calculate the compound score of the sentences. Each compound score of the
sentences was translated into a readable format with a score of over 0.05 labeled as ’positive,’ a
score below -0.05 as ’negative,’ and a score in between as ’neutral.’
Within the contemporary field of Sentiment Analysis, lexicon-based methods have garnered
a degree of criticism in regards to their general limitation in ”accuracy” and ”validation”
as compared to data-driven methods. In prior Sentiment Analysis research, many found that
lexicon-based methods’ have lower efficacy when validating results with human annotators
[¨
Ohman,2021,Teodorescu and Mohammad,2022].
This phenomenon is well documented with several lexicon-based packages, such as R’s Syuzhet
[Jockers,2017], which prioritizes word frequency to its analysis (leading to misclassification
of sentiments to text) alongside their limited ability to detect negation. Nonetheless, studies
have also shown that lexicon-based methods can still be a valid tool for analyzing the emotions
and sentiments within texts. Research comparing the efficacy of VADER and TextBlob Lo-
ria et al. [2018] (another lexicon-based analysis tool) using tweets of COVID vaccines found
that VADER classification generally corresponded better than TextBlob when directly compar-
ing human annotation results, while another study analyzing tweets of the 2016 US election
(exclusively using VADER) yielded satisfactory results of accuracy according to its authors
[Georgios-Alexandros,2022,Elgagir and Yang,2019]. Therefore, along with the benefit of
low-cost implementation (i.e., the negation of classification by human annotators for machine
learning models), lexicon-based methods, including VADER, are capable of providing useful
and even desirable insights for researchers that are comparable to machine learning methods
[¨
Ohman,2021]. Moreover, ¨
Ohman [2021] suggests that the lexicon-based approach has its
merits when the purpose of the study is not to achieve higher numerical accuracy in sentiment
classification but to focus on investigating emotion-associated words.
Therefore, the application of VADER is a useful and justified method in this study, as the goal
is to provide a generalized overview of positive and negative contents (via sentences) within the
manifestos and attempt to understand the trend of emotion usage while being able to maintain
a satisfying level of accuracy compared to other similar lexicon-based packages.
3.2 Method 2: NRC Emotion lexicons
Along with VADER, NRC Word-Emotion Associations (EmoLex), another lexicon-based emo-
tion detection and analysis method was used. EmoLex was created via crowd-sourcing to as-
sociate existing English lexicons with various emotions by manual human annotation [Moham-
mad and Turney,2012]. Unlike VADER, which is only capable of ternary sentiment classifi-
cations, EmoLex annotators have additionally assigned each lexicon with the eight basic affect
categories: joy, trust, anticipation, surprise, fear, sadness, anger, and disgust [Mohammad and
Turney,2012]. In effect, EmoLex is capable of analyzing the deeper dynamics of specific affect
emotion-laden lexicons and their distribution beyond the limited binary positive and negative
sentiment classification, a task considered the most suitable for lexicon-based methods and in-
corporated into a previous study on post-war Finnish manifestos [ ¨
Ohman,2021,Koljonen et al.,
2022]. Furthermore, ¨
Ohman [2021] proposes that emotion or sentiment analysis research using
lexicon-based models should be evaluated by sanity checks to verify the usefulness and accu-
racy of any given results. As such, EmoLex also aims to act as an evaluation of the VADER
results (and vice versa) to check the rationality and validity of the results obtained.
While the deployment of EmoLex was largely similar to that of VADER, we further included
the lemmatization of sentences and the removal of punctuation and other non-alphanumeric
Journal of Data Mining and Digital Humanities
ISSN 2416-5999, an open-access journal
4http://jdmdh.episciences.org
characters. The processed sentences were later merged as a single text used for EmoLex analy-
sis. In this study, the affect frequency, which returns the frequency with the range value of 0 to
1 for the two sentiments and eight affect emotions from the analyzed manifesto text, was used
to visualize the emotion-word frequencies.
IV RESULTS
4.1 VADER Sentiment Analysis Results
The results of VADER sentiment analysis of both parties’ manifestos are shown in table 1and
table 2, respectively.
Year Sentences GOV Status Pos share(%) Pos Change Neg share(%) Neg Change Neut share(%)
2001 977 Incumbent 62.641 0.0 16.07 0.0 21.29
2005 801 Incumbent 63.92 1.28 19.725 3.7 16.355
2010 1313 Incumbent 66.565 2.64 16.375 -3.4 17.06
2015 865 Opposition 60.231 -6.33 22.89 6.5 16.879
2017 1136 Opposition 53.257 -6.97 24.032 1.1 22.711
2019 1192 Opposition 50.336 -2.92 30.369 6.3 19.295
Table 1: Sentiment Results of Labour Party Manifesto sentences (2001 - 2019)
Year Sentences GOV Status Pos share(%) Pos Change Neg share(%) Neg Change Neut share(%)
2001 680 Opposition 48.382 0.0 25.735 0.0 25.882
2005 415 Opposition 53.012 4.63 20.241 -5.5 26.747
2010 494 Opposition 58.3 5.29 22.065 1.8 19.636
2015 283 Incumbent 75.618 17.32 13.428 -8.6 10.954
2017 1275 Incumbent 70.667 -4.95 13.961 0.5 15.373
2019 974 Incumbent 64.682 -5.98 15.811 1.8 19.507
Table 2: Sentiment Results of Conservative Party Manifesto sentences (2001 - 2019)
The results in table 2for the Labour Party show that sentences with positive sentiments com-
promised over 60% of the total share of sentences within the manifestos. This is seen in the
Pos share (percentage share of positive sentences out of the total) during its incumbency, with
its Pos Change (the change in shares of positive sentences from the previous election) showing
a small increase in share from 2001 to 2010. These 3 (2001, 2005, and 2010) Pos share figures
were each higher than those of Conservative Party Manifestos published during the correspond-
ing election year as presented in table 2. However, after the 2010 election, when the Labour
Party was relegated to the opposition, there appears to be a gradual drop in Labour Party’s
Pos share from 2015 to 2019, with the Pos share of the Labour Party’s last manifesto in 2019
being only around 50%. At the same time, post-2010 Labour manifestos saw an increase in total
shares of negative sentences (Neg share), further seen by the highest Neg Change (the change
in total shares of negative sentences from the previous election) of over 6 points in 2015 and
2019. This has led the Neg share of the Labour Party’s 2019 manifesto to be 30%, doubling the
Neg share of the Labour Party’s 2001 manifesto. Additionally, as the opposition, the Neg share
of all Labour Party manifestos in the last three elections (2015, 2017, and 2019) was larger than
all the Conservative Party manifestos published in the same election cycle.
The results of the Conservative Party manifestos are presented in table 2. As mentioned before,
during their time as the opposition from 2001 to 2010, the Pos share of the Conservative Party’s
manifesto was always smaller than those of the Labour Party manifestos. This is not the case
for Neg share, which, for the first three elections of 2001, 2005, and 2010, the Conservative
Journal of Data Mining and Digital Humanities
ISSN 2416-5999, an open-access journal
5http://jdmdh.episciences.org
Party manifestos had a higher share of negative sentences than the Labour Party. This trend
changed after the Conservative Party returned to power in the 2010 elections, becoming the
incumbent. Starting from the 2015 election, the Conservative Party’s Pos share immediately
overtook the Labour Party. Specifically, the share of positive sentences in the Conservative
Party’s 2015 manifesto saw a 17-point increase in Pos Change from 2010 and a 15-point lead
against Labour’s manifesto published in the same year. While their Pos Change in 2017 and
2019 turned negative (indicating the decline of total shares in positive sentences in Conservative
manifestos), the Conservative Party’s Pos share were much higher than the Labour Party in the
last two elections.
4.2 EmoLex Emotion Analysis Results
Moving to the EmoLex results, Appendix 1 presents the 6 line graphs representing the frequen-
cies of Positive/Negative sentiments and the 8 affect emotions via TJA and FASD graphs for the
Labour Party and the Conservative Party manifestos. (Note: The TJA Graph represents posi-
tively associated emotions: Trust, Joy, and Anticipation. The FASD Graph represents negatively
associated emotions: Fear, Anger, Sadness, and Disgust)
Starting with the Labour Party’s graph for the frequency of Positive and Negative lexicons, dur-
ing the 2001-2010 period, positive lexicons had a frequency of 0.3 while negative lexicons were
just 0.1. Comparing this to the graph for the Conservative Party, we find that the frequency of
the Labour Party for positive lexicons was higher in the first three elections during this period.
In return, however, the Conservative Party had a higher frequency in the negative lexicon. This
pattern, however, isn’t reflected in the TJA and FASD Graphs. For example, while the Labour
Party had a lead in the frequency of Trust lexicons in the TJA graph against the Conservative
Party in the 2001 election, the Conservative Party had overtaken them in the following elec-
tion in 2005 despite being the opposition. Moreover, neither party had a constant lead in the
frequency of fear, anger, and sadness during this period.
Since the 2010 elections, however, a salient pattern has emerged where the Labour Party man-
ifestos began showing an increase, eventually taking the lead in the frequency of negative and
negatively associated FASD lexicons in the 2019 election. Specifically, the frequency of neg-
ative lexicons in the Labour Party’s 2010 manifesto was initially 0.1; this increased to around
0.14 by 2019. Regarding the lexicons related to fear, it also saw an increase starting from 0.05
in 2010 to below 0.08 at the same time by 2019. In comparison, the frequency of negative
and FASD lexicons within the Conservative Party manifestos shows a minor decline over time.
In its replacement, positive and positively associated TJA lexicons are more frequently used
in Conservative Party manifestos over time. And by the 2019 election, the Conservative Party
used positive and positively associated TJA lexicons more frequently than the Labour Party.
Journal of Data Mining and Digital Humanities
ISSN 2416-5999, an open-access journal
6http://jdmdh.episciences.org
Figure 1and Figure 2, represent the combined result of the 3 line graphs given to each party.
Figure 1: Labour party change in emotion-associated content
Figure 2: Conservative party change in emotion-associated content
Combining all the EmoLex results into a single graph (Figure 1for the Labour Party and Figure
2for the Conservative Party), we find that positive languages (especially positive sentiments
and the emotion of trust) are most frequently adopted for both parties’ manifestos regardless of
their position as opposition or incumbent. Nevertheless, there is a notable trend in the Labour
manifestos in Figure 1where negative emotions gradually replace the share of frequency of
certain positive emotions. This is particularly highlighted in the 2019 manifesto, where the
frequency of negative lexicons was closely behind trust lexicons for the Labour Party.
Figure 3is a heat map created from both parties’ combined EmoLex results to identify the
correlation between the frequency of sentiment and affect emotions against the party’s status
in government (as incumbent or opposition). The results indicate that positive emotions are
relatively correlated with the party’s status in government (GOV Status); where a party becomes
the incumbent, positive language will likely increase. On the other hand, negative emotions are
negatively correlated, indicating that they will increase when the party becomes the opposition.
Journal of Data Mining and Digital Humanities
ISSN 2416-5999, an open-access journal
7http://jdmdh.episciences.org
Figure 3: Correlation matrix for both parties combined
The results summarized in this section will be further analyzed with the hypotheses addressed
in section 2.
V ANALYSIS OF RESULTS
Assessment of H1: The frequency of positive emotions will be highest in the incumbent
party’s manifesto than the opposition, while the frequency of negative emotions will be high-
est in the opposition party’s manifesto than the incumbent
The results presented in the VADER sentiment analysis and EmoLex effectively support the
assumptions of the first hypothesis (H1). This is presented in figure 3, where we find a strong
correlation between government status and positive/negative lexicons. The findings are fur-
ther supported by the VADER results exhibited in table 1and table 2. Based on these results,
the incumbent Conservative Party had more total shares of positive sentences in all post-2010
elections than the opposition Labour Party. However, as the opposition, the Labour Party had
more shares of negative sentences than the Conservative Party. The pattern presented in the
results from VADER is also reciprocated in the EmoLex results in the post-2010 elections. Ac-
cording to Appendix 1 and figure 1, after being relegated to the opposition, the Labour Party
had increased the frequency of negative lexicons in their manifestos, eventually surpassing the
Conservative Party. During the same period, the Conservative Party, now the incumbent, began
increasing the frequency of positive lexicons and later surpassing the Labour Party manifestos
as demonstrated in figure 2.
Both findings for VADER and EmoLex are in line with previous research, in which incumbent
parties, due to their increased responsibility of governance associated by the public, would use
Journal of Data Mining and Digital Humanities
ISSN 2416-5999, an open-access journal
8http://jdmdh.episciences.org
positive language at a higher rate to improve their image towards the public [Crabtree et al.,
2018]. At the same time, the opposition party would use more negative language to criticize
and tarnish the incumbent(s), such as pointing to their failure of governance. This pattern ar-
guably manifested in UK general election manifestos of 2017 and 2019 following the polarizing
2016 Brexit referendum. Given that the Conservative Party was largely held responsible by the
British public to guide the country after the Brexit decision, their 2017 and 2019 manifesto
needed to include positively reinforcing messages to boost public confidence in their govern-
ment [Bonnet,2021]. In contrast, the Labour Party used its status as the opposition to scrutinize
the Conservative government, including their ’mishandling’ of Brexit and critically opposing
their public policies [Hayton,2021]. These dynamics have arguably led the Conservative mani-
festos to increase and maintain a high frequency of positive emotions within their manifestos. In
contrast, the Labour Party manifestos displayed a higher frequency of negative emotions after
the 2010 election, validating H1.
Assessment of H2: Manifestos of ideologically similar parties will use positive language more
frequently than negative language
The VADER and EmoLex results also show that positive languages had the highest frequency
in both parties across all elections, regardless of the party’s position as incumbent or opposi-
tion. This supports H2, where parties with ideological similarities would largely compete with
positive language in their manifestos. The Pos share of table 1and table 2show that sentences
with positive sentiments constitute at least half of the total sentences within the manifesto. In
contrast, Neg share rarely constituted a third of the total. Additionally, figure 1and figure 3
show that positive lexicons (excluding TJA lexicons) accounted for the highest frequency, as
predicted in previous research relating to H2.
Despite the results, when comparing the Conservative Party’s EmoLex frequencies, the Labour
Party resorted to increasing the frequency of negative lexicons (including FASD emotions) and
even displacing several positively associated TJA lexicons in frequency by 2019. While these
trends can partially be explained through the analysis of H1, another factor that may have influ-
enced Labour Party manifestos to increase the usage of negative lexicons, especially in 2017 and
2019, is also the left-wing shift of the Labour Party under the leadership of Jeremy Corbyn from
2015 to 2020 [Manwaring,2019]. Under Corbyn, many had remarked on the Labour Party’s
increasing return to advocating traditional left-wing economic and social policies, thereby shift-
ing the party’s ideology to the left and away from the past centrist positions, leading some to
describe this move as ’radicalization’ of the party. While some question the extent of the Labour
Party’s ideological shifting under Corbyn’s leadership, close research on policies presented in
Labour 2017 and 2019 manifesto demonstrates a clear trend of left-leaning shift [Manwaring,
2019,Jacobs and Hindmoor,2022].
Referring to the study by Kosmidis et al. [2018], parties with similar ideologies and policy
positions often rely on positive ”emotional appeals” during their campaign to compete over the
same electorate. Radical fringe parties, however, would commonly adopt negative languages to
mobilize disaffected voters, as suggested by Crabtree et al. [2018]. This may help to explain
why negative and FASD lexicons overtake certain TJA lexicons in the Labour Party’s manifesto
from 2017 to 2019, as seen in Figure 1, as it likely represents the growing radicalization of the
Labour Party, which increasingly promoted radical policies compared to the Conservative Party,
impacting their amount of positive lexicons as explained by Crabtree et al. [2018].
On the surface, results for VADER and EmoLex support H2, with positive languages holding the
Journal of Data Mining and Digital Humanities
ISSN 2416-5999, an open-access journal
9http://jdmdh.episciences.org
highest frequency through all elections in both the Labour Party and the Conservative Party re-
gardless of their position as incumbent or opposition. However, the critical rise in the frequency
of negative emotions in the Labour Party’s 2017 and 2019 manifesto likely reflects its political
radicalization and partial ideological divergence from the Conservative Party. Nonetheless, the
results can be observed as evidence of how ideological positioning can factor in the dynamics
of emotional language within manifestos.
VI CONCLUSION
The primary aim of this research was to investigate the impact of a party’s current government
status and position of ideology on the use of emotional language in their election manifesto.
Therefore, using the recent UK election manifestos as data, we tested the two hypotheses to
investigate this relationship. To summarize the findings, VADER and EmoLex analysis under-
scores the strong influence a party’s incumbent or opposition status and ideological positioning
have on the frequency of positive and negative language in manifestos.
In future work, we hope to address several limitations, including preprocessing of texts, adding
domain-specific terms to the emotion lexicons, and incorporating additional manifesto data. As
it currently stands, the vastly different formats of the manifestos might have introduced some
artefacts in the text extraction process, of which the majority were manually removed. However,
some may remain.
Additionally, since the EmoLex was originally annotated using North American annotators,
some cultural differences in emotion associations might exist. Furthermore, the domain of the
manifestos is very specific, and likely, several terms were not properly included. In future
studies, the lexicon should be made more domain-specific.
Finally, as the manifesto data is limited to those produced since the 2000s, it only presents a
period when both the Labour Party and the Conservative Party already displayed strong similar-
ities in ideology. While H2 aimed to investigate the impact this had on the emotive frequency
in manifestos, future studies should incorporate additional data from previous elections when
both parties presented far stronger ideological and policy distinctions to further test the validity
of our hypotheses.
References
Loris Belcastro, Riccardo Cantini, Fabrizio Marozzo, Domenico Talia, and Paolo Trunfio. Learning political
polarization on social media using neural networks. IEEE Access, 8:47177–47187, 2020. doi: 10.1109/access.
2020.2978950.
Steven Bird, Ewan Klein, and Edward Loper. Natural language processing with Python: analyzing text with the
natural language toolkit. ” O’Reilly Media, Inc.”, 2009.
Vernon Bogdanor. General election: Do party manifestos still matter?, Apr 2015. URL https://www.ft.
com/content/7418b410-e2c2-11e4-bf4b-00144feab7de.
Alma-Pierre Bonnet. Dealing with the crisis: The brexit spin. Observatoire de la soci´
et´
e britannique, (27):
103–125, 2021. doi: 10.4000/osb.5365. URL https://journals.openedition.org/osb/5365.
Pio Calderon. Vader sentiment analysis explained, Mar 2018. URL https://medium.com/
@piocalderon/vader-sentiment-analysis-explained-f1c4f9101cd9#:˜:text=
VADER%20(Valence%20Aware%20Dictionary%20for,intensity%20(strength)%20of%
20emotion.
Charles Crabtree, Matt Golder, Thomas Gschwend, and Indridi H. Indridason. It’s not only what you say, it’s also
how you say it: The strategic use of campaign sentiment. 2018. doi: 10.31235/osf.io/g2sd6.
Nikolaus Eder, Marcelo Jenny, and Wolfgang C. M¨
uller. Manifesto functions: How party candidates view and use
their party’s central policy document. Electoral Studies, 45:75–87, 2017. doi: 10.1016/j.electstud.2016.11.011.
Journal of Data Mining and Digital Humanities
ISSN 2416-5999, an open-access journal
10 http://jdmdh.episciences.org
Shihab Elgagir and Jing Yang. Twitter sentiment analysis using natural language toolkit and vader sentiment.
Proceedings of the International MultiConference of Engineers and Computer Scientists 2019, page 12–16,
2019. URL https://www.iaeng.org/publication/IMECS2019/IMECS2019_pp12-16.pdf.
Patrick English, Maria T. Grasso, Barbara Buraczynska, Sotirios Karampampas, and Luke Temple. Convergence
on crisis? comparing labour and conservative party framing of the economic crisis in britain, 2008-14. Politics
and Policy, 44(3):577–603, 2016. doi: 10.1111/polp.12160.
Asderis Georgios-Alexandros. Sentiment analysis on Twitter data: a detailed comparison of TextBlob and VADER.
PhD thesis, Jun 2022. URL https://repository.ihu.edu.gr//xmlui/handle/11544/29947.
Richard Hayton. Brexit and party change: The conservatives and labour at westminster. International Political
Science Review, 43(3):345–358, 2021. doi: 10.1177/01925121211003787.
Clayton Hutto and Eric Gilbert. Vader: A parsimonious rule-based model for sentiment analysis of social media
text. In Proceedings of the international AAAI conference on web and social media, volume 8, pages 216–225,
2014.
Michael Jacobs and Andrew Hindmoor. Labour, left and right: On party positioning and policy reasoning. The
British Journal of Politics and International Relations, 2022. doi: 10.1177/13691481221099734. URL https:
//eprints.whiterose.ac.uk/185728/.
Carsten Jentsch, Enno Mammen, Henrik M¨
uller, Jonas Rieger, and Christof Sch¨
otz. Text mining methods for mea-
suring the coherence of party manifestos for the german federal elections from 1990 to 2021. DoCMA Working
Papers, Sep 2021. doi: 10.17877/de290r-22363. URL https://doi.org/10.17877/de290r-22363.
Matthew Jockers. Package ‘syuzhet’. URL: https://cran. r-project. org/web/packages/syuzhet, 2017.
Juha Koljonen, Emily ¨
Ohman, Pertti Ahonen, and Mikko Mattila. Strategic sentiments and emotions in post-
second world war party manifestos in finland. Journal of Computational Social Science, 5(2):1529–1554, 2022.
doi: 10.1007/s42001-022-00181-9.
Spyros Kosmidis, Sara B. Hobolt, Eamonn Molloy, and Stephen Whitefield. Party competition and emotive
rhetoric. Comparative Political Studies, 52(6):811–837, 2018. doi: 10.1177/0010414018797942.
Steven Loria et al. textblob documentation. Release 0.15, 2(8), 2018.
Tom Louwerse, Ulrich Sieberer, Or Tuttnauer, and Rudy B. Andeweg. Opposition in times of crisis: Covid-19
in parliamentary debates. West European Politics, 44(5-6):1025–1051, 2021. doi: 10.1080/01402382.2021.
1886519.
Rob Manwaring. Corbyn’s labour agenda has more in common with its forbears than is often assumed, May 2019.
URL https://blogs.lse.ac.uk/politicsandpolicy/corbyns-labour-agenda/.
Saif M. Mohammad and Peter D. Turney. Crowdsourcing a word-emotion association lexicon. Computational
Intelligence, 29(3):436–465, 2012. doi: 10.1111/j.1467-8640.2012.00460.x.
Emily ¨
Ohman. The validity of lexicon-based sentiment analysis in interdisciplinary research. In Proceedings of the
Workshop on Natural Language Processing for Digital Humanities, pages 7–12, NIT Silchar, India, December
2021. NLP Association of India (NLPAI). URL https://aclanthology.org/2021.nlp4dh-1.2.
Daniela Teodorescu and Saif M Mohammad. Frustratingly easy sentiment analysis of text streams: Generating
high-quality emotion arcs using emotion lexicons. arXiv preprint arXiv:2210.07381, 2022.
Wikipedia. List of conservative party (uk) general election manifestos, Nov 2022a. URL https:
//en.wikipedia.org/wiki/List_of_Conservative_Party_(UK)_general_election_
manifestos.
Wikipedia. List of labour party (uk) general election manifestos, Dec 2022b. URL https://en.wikipedia.
org/wiki/List_of_Labour_Party_(UK)_general_election_manifestos.
Pinar Yildirim. How social media is shaping political campaigns, Aug 2020. URL https:
//knowledge.wharton.upenn.edu/podcast/knowledge-at-wharton-podcast/
how-social-media-is-shaping-political-campaigns/.
Journal of Data Mining and Digital Humanities
ISSN 2416-5999, an open-access journal
11 http://jdmdh.episciences.org
