2.1. Social Media Platforms and Digital Political Engagement

Social media platforms have reshaped political engagement by directly enabling emotionally charged interactions between candidates and voters. Platforms such as Facebook, X, and TikTok are no longer passive information outlets. They function as an active space where political sentiment is expressed, shaped, and amplified in real time (Boulianne, 2020; Persily & Tucker, 2020).

Emotional reactions have emerged as a key metric of user sentiment on social media platforms. Tian et al. (2017) showed that Facebook emoji reactions offer interpretable affective signals, which are often more reliable than text-based sentiment. Bond et al. (2012) found that emotionally resonant content on Facebook could influence real-world voter turnout, underscoring the behavioral power of digital emotion. Several studies link social media sentiment to electoral outcomes. Tumasjan et al. (2010) and Haryanto et al. (2019) demonstrated how X and Facebook sentiment mirrored election results in Germany and Indonesia, respectively. Muraoka et al. (2021) found that love and anger reactions to party posts correlated with public support across 79 democracies. These findings suggest that emoji reactions are more than mere expressions, as they serve as behavioral signals with predictive potential.

In the context of South Asia, Agarwal and Bansal (2020) examined Facebook posts during India’s 2017 state election and found that sentiment polarity correlated with candidate performance. In Nepal, Bhattarai (2023) noted Facebook’s growing role in political socialization, particularly among youth, while Upreti et al. (2025) found that emotional engagement on Facebook correlated with electoral success in the 2022 parliamentary election. Their study revealed patterns in voter sentiment, messaging strategies, and emotional tone. However, challenges including misinformation and echo chambers can skew sentiment (Guess et al., 2020), and in Nepal, limited digital access may affect representativeness (Kharel, 2024).

Despite this emerging body of work, comprehensive analyses of Facebook-based affective signaling in Nepal remain limited. In particular, the role of emoji reactions and emotional cues as behavioral indicators of political alignment remains underexplored, highlighting the need for further research into AIB in the country’s digital political sphere.

2.2. Emotion Theory and Affective Information Behavior

Understanding how emotions shape information processing is crucial for interpreting digital political behavior. For classifying emotional responses, emotion theory provides the psychological foundation, while AIB explains how these emotional states shape the way individuals interact with information in real-world and digital environments.

Ekman (1993)’s theory of six universal emotions (happiness, sadness, fear, anger, surprise, and disgust) offers a widely accepted framework for decoding affective expression. Similarly, Plutchik (1965)’s wheel of emotions illustrates how primary emotions blend into complex affective states that influence perception, communication, and decision-making. In digital spaces like Facebook, emoji reactions such as love, care, anger, or sadness serve as proxies for these emotional expressions, allowing users to communicate affective responses to political content on a scale.

To further explore this emotional analysis with information science, this study draws from the field of AIB, which explores how emotional states influence information seeking, processing, and use. Kuhlthau (1991)’s information search process (ISP) model was among the first to integrate affect, identifying phases of uncertainty, optimism, confusion, and relief during search behavior. Nahl (2007) later expanded this with a tripartite model incorporating affective, cognitive, and sensorimotor dimensions. Savolainen (2015) emphasized the everyday, contextual nature of affective engagement, where emotions guide not only what information individuals attend to, but also how they evaluate and act upon it.

In this study, Facebook emoji reactions are conceptualized not just as affective signals but as manifestations of collective emotional information processing. Users reacting to political posts on social media are engaging in a form of public affective behavior—an expression of sentiment that reflects not only individual preferences but also broader political association. By aggregating and analyzing millions of such reactions, we argue that these aggregated affective signals function as behavioral indicators, offering predictive insight into electoral outcomes. This integration of emotion theory with AIB thus enables a more holistic understanding of how affect drives both digital engagement and real-world political decision-making.

4.1. Data Collection and Sample

We collected data from the official Facebook pages of three major mayoral candidates in the 2022 Kathmandu metropolitan election: Balen Shah (independent), Keshav Sthapit (CPN-UML), and Sirjana Singh (NC). These mayoral candidates were selected due to their prominence and public engagement throughout the campaign. Given the capital city’s central political role and strong digital infrastructure, the election attracted a highly engaged online audience.

Table 1 summarizes the demographic and politician background of the three main candidates in the 2022 Kathmandu mayoral election. Balen Shah, a 35-year-old independent candidate and structural engineer with a background in rap music, appealed strongly to younger, digitally active voters through his outsider image and reform-oriented message. In contrast, Keshav Sthapit, a 69-year-old seasoned politician from the CPN-UML with prior experience as mayor (1997-2006), drew support from older, traditional party members. Sirjana Singh, aged 63 and representing the NC, came from a well-established political lineage and was supported primarily by unions and party-affiliated constituents.

Earlier, a national newspaper conducted an opinion poll in the days before the election (Setopati Team, 2020). The poll results showed that Shah, the independent candidate, was in the lead, followed by Sthapit from the CPN-UML party in second place, and Singh from the NC party in third place. The poll also revealed minimal support for other candidates, reflecting the preferences of voters who had not decided yet (Fig. 2).

Fig. 2

Opinion poll for mayoral candidates in percentages.

To mimic real-time predictive analysis, we retrieved Facebook posts, reactions, comments, and shares from April 26 to May 10, 2022, excluding the final three days to avoid post hoc bias. Preprocessing included the removal of spam, duplicates, and non-emotive content to ensure a high-quality dataset for sentiment evaluation.

4.2. Sentiment Categorization and Scoring

The initial task involved categorizing Facebook emojis into five impressions: very positive, positive, neutral, negative, and very negative expressions (Table 2), drawing on the emotional models identified by Ekman (1993) and Plutchik (1965), which primarily delineate three types of emotions. We assigned a weighted sentiment score to each reaction type based on its implied emotional value into positive (e.g., love, care, and like), negative (e.g., sad and angry), and neutral (e.g., ha-ha and wow) reactions. These emojis were deemed universally applicable to any text, audio, video post, or share on the platform. To further ensure reliability, we calculated the weights using a pilot study with human annotators, achieving an inter-rater agreement.

Users typically employ the ‘like’ emoji when expressing agreement and positive sentiment on a post. The ‘care’ reaction is used to convey empathy, kindness, and understanding, and it is also inclined towards positive feelings. Expressing greater affection, users utilize the ‘love’ emoji, considered twice as strong as the like and care emoji, to ensure an accurate calculation of positive polarity on Facebook.

On the other hand, ‘sad’ and ‘angry’ reactions were used to express disapproval or frustration. Given its heightened intensity, the ‘angry’ emoji was assigned double the weight of ‘sad.’ Emojis such as ‘ha-ha’ and ‘wow’ were treated as context-dependent: ‘ha-ha,’ often used sarcastically in political discourse, was classified as negative, while wow was considered neutral to mildly positive, particularly when found in combination with other favorable reactions. Utilizing these data, the overall positive and negative sentiments were calculated accordingly as follows:

Table 3 illustrates the sum of all seven emojis, including followers, posts, comments, and shares. Balen Shah had 30 posts, Keshav Sthapit had 168 posts, and Sirjana Singh had 124 posts on their respective Facebook walls. The cumulative number of posts amounted to 322, and all the emotions and reactions, including followers associated with each post and candidates, were meticulously analyzed.

Using the first formula and data from Table 3, we calculate the total positive and negative sentiments in Table 4 below.

Based on the data of total positive and negative sentiment, each post’s net sentiment polarity was calculated as:

Sentiment polarity

4.3. Predictive Signal Assessment

To evaluate the predictive potential of affective engagement of Facebook, we employed a combination of interpretable, signal-based approaches grounded in sentiment metrics and pre-election data. These included:

(i) Rank-order comparison between digital sentiment metrics and vote shares.

(ii) Pearson correlation to quantify the strength of association.

Key indicators such as average sentiment polarity, engagement rate (e.g., reactions, shares, comments per 1,000 followers), and overall emotional reaction scores were computed for each candidate and compared against their final vote percentage. To further strengthen this analysis, we applied the logarithmic sentiment index developed by Oh and Kumar (2017), which expresses the ratio of positive to negative reactions:

This framework enabled a focused assessment of whether Facebook-based affective expressions collected before elections could meaningfully signal voter support. The strong alignment between sentiment indicators and electoral outcomes highlights the potential of social media data for early-stage electoral forecasting and behavioral analysis in digital political environments.

5.1. Findings from Emotions and Engagement

Table 5 reveals great contrasts in both emotional intensity and engagement patterns across the three mayoral candidates. Balen Shah has an exceptionally high emotion score of 1,429,539, compared to just 17,450 and 16,021 for Sirjana Singh and Keshav Sthapit, respectively. This shows many folds of difference, suggesting that Shah not only generated more interactions, but also evoked far stronger emotional responses among Facebook users. His posts elicited highly affective reactions, indicating that followers were not merely passive consumers of content but deeply engaged participants.

In terms of engagement rate and reactions per 1,000 followers, Shah again dominated, with values of 980 outpacing Singh (355) and Sthapit (243) by fair margins. This indicates that his emotional appeal translated into active engagement on a per-follower basis, further reinforcing the notion of authentic resonance between candidate and audience.

Notably, Shah also led in shares per post (1,329) and comments per post (2,701), both orders of magnitude higher than Singh (42.8 shares, 73.6 comments) and Sthapit (12.8 shares, 39.5 comments). High sharing rates suggest that Shah’s content was not only engaging but amplified organically through network effects, reaching a broader audience and reinforcing campaign visibility. The volume of comments further reflects a high level of public discourse and two-way interaction surrounding his candidacy.

Together, these metrics reflect a candidate whose campaign successfully leveraged affective communication and participatory momentum on Facebook. The scale and depth of digital engagement observed for Shah support the broader finding that emotion-driven social media activity is a meaningful predictor of electoral support.

Similarly, Fig. 3 shows the average Facebook engagement (likes, shares, and comments) for each candidate, calculated over two-day intervals from April 26 to May 10, 2022. Balen Shah began his campaign on April 28 but quickly gained traction, consistently leading with significantly higher engagement, peaking at nearly 120,000 on May 4 before declining later. In contrast, Sthapit and Singh showed relatively low and stable engagement, with minor fluctuations. The stark contrast highlights Shah’s strong digital presence and sustained public interest during the campaign period, aligning with his eventual electoral success.

Fig. 3

Daily Facebook engagement trend.

5.2. Findings from Predictive Signals

To assess the predictive strength of Facebook reactions, we compare each candidate’s pre-election sentiment indicators with their actual vote share.

A rank-order comparison was conducted to evaluate the alignment between pre-election sentiment metrics and actual electoral outcomes based on both average sentiment polarity and the logarithmic sentiment index. Balen Shah consistently ranked first across all three dimensions—average polarity (1.00), sentiment index (6.84), and vote share (38.64%), demonstrating a strong alignment between emotional engagement and electoral success. Although Singh and Sthapit’s vote shares were nearly identical (23.83% and 24.03%, respectively), their positions were reversed compared to their sentiment ranks (Table 6). Despite this minor discrepancy, the overall ranking demonstrates strong alignment between affective digital engagement and electoral performance, reinforcing the predictive value of pre-election Facebook sentiment signals.

To further evaluate the relationship between online sentiment and electoral outcomes, a Pearson correlation was conducted using both the sentiment index and net sentiment polarity. The correlation between the log-transformed sentiment index and vote share was r=0.81, indicating a strong positive association. This metric reflects the relative dominance of positive sentiment over negative. A moderate correlation was observed between net sentiment polarity and vote share (r=0.58), suggesting that the overall emotional tone of user reactions (particularly the balance between positive and negative sentiment) was meaningfully aligned with electoral performance. This was evident in Shah’s case, where high user engagement and affective support corresponded closely with voting outcomes.

These findings suggest that Facebook sentiment signals (net polarity and sentiment index) offer a reliable behavioral proxy for electoral support when measured prior to elections. However, minor rank mismatches between sentiments and vote share, as seen between Singh and Sthapit, indicate that offline factors such as party affiliation, demographic outreach, or mobilization strategies may still influence voting outcomes beyond what is captured digitally.

5.3. Robustness Checks

To ensure the reliability of our findings, we conducted two robust validation tests. First, we conducted a time-truncation analysis by excluding the final three days of Facebook data (May 8-10) from the dataset. After truncating, 201 posts remained across all candidates. Keshav Sthapit had the most posts, followed by Sirjana Singh and Balen Shah, consistent with the post distribution in the full dataset. We then ran the sentiment analysis on the truncated dataset and compared the results with the actual vote share. While the correlation between sentiment analysis and vote share declined slightly (r=0.78), the ranking of candidates and the overall predictive direction remain unchanged. This suggests that the model’s predictive strength is not solely dependent on last-minute surges in online engagement but instead reflects more stable affective patterns throughout the campaign (Jaidka et al., 2019).

Second, we tested the robustness of our sentiment scoring method by removing the emotion-based weighting assigned to specific emoji reactions. In this unweighted model, all reactions were treated equally, so each emoji contributed to one unit regardless of its affective intensity. While this simplified approach still yielded a strong correlation between sentiment polarity and vote share (r=0.62), the predictive precision declined slightly compared to the weighted model. This result reinforces the theoretical and empirical value of emotion-weighted scoring, which captures varying degrees of affective expression and improves the sensitivity of the sentiment-vote relationship (Giuntini et al., 2019; Tian et al., 2017).

These checks confirm that our results are not dependent on a specific time window and emoji weighting scheme, strengthening the claim that Facebook emoji reactions are robust behavioral signals of electoral sentiment (Hauthal et al., 2019).

5.4. Findings from Text Analytics

We conducted text analytics of Facebook comments directed toward mayoral candidates Balen Shah, Keshav Sthapit, and Sirjana Singh to uncover the most frequently mentioned messages, terms, and sentiments associated with each. Using techniques such as word clouds and sentiment analysis, we gained valuable insights into the narratives and conversations of the electorate. Word clouds offer a visually intuitive method for highlighting commonly used terms, making them crucial for identifying key public conversations.

We began by analyzing the word cloud for Balen, who received the most comments—76,141 in total, which is 85% more than the combined total for the other two candidates. Most of these comments were positive, with phrases like ‘I support Balen,’ ‘my vote for Balen,’ and ‘time for change’ appearing frequently (Fig. 4). Additionally, many comments from users included terms like ‘youth’ and similar phrases, such as ‘vote for youth,’ ‘youth for change,’ ‘youths of Kathmandu,’ and ‘Balen rock.’ Balen’s posts primarily featured photos from campaign rallies and live videos. He was the first to announce his mayoral candidacy and start a Facebook campaign. Judging by the comments and rally videos, it appears that he had many young supporters. On the negative side, Shah was underestimated for his independent candidacy without the support of any political party, which is a significantly necessary factor in Nepal’s partisan political landscape.

Fig. 4

Word clouds for Balen Shah (top left), Keshav Sthapit (top right), and Sirjana Singh (bottom).

In the word clouds for Keshav Sthapit and Sirjana Singh, we observed a noticeable concentration of support for their rival, Shah, with comments like ‘vote for Balen’ and ‘this time it is Balen’ appearing occasionally. Despite this, there were numerous positive comments for both candidates. Keshav Sthapit’s posts received comments such as ‘good luck,’ ‘congratulations,’ and ‘be the best again,’ reflecting his previous tenure as mayor two decades earlier from 1997 to 2006, as he sought this position for a second time. Similarly, Sirjana Singh’s posts attracted comments including ‘best wishes,’ ‘Jay (hail) Nepal,’ ‘congratulations,’ and ‘gentlewoman.’ However, despite support from various unions, including those representing teachers, students, traders, and drivers, both candidates also faced substantial criticism. Keshav Sthapit was criticized with comments like ‘please talk less’ and ‘you are not eligible,’ while Singh encountered remarks such as ‘no nepotism,’ and ‘shame to Ganesh Man.’ Mrs. Singh comes from a prominent political background, with her husband serving as a parliamentarian and her father-in-law, Ganesh Man, a significant figure in Nepal’s political history. She was often criticized for alleged candidacy bargaining and partisan politics. Additionally, users who expressed support were derogatorily referred to as ‘jholye’ (porter) and ‘bheda’ (sheep). The term ‘jholye’ is used to describe people who blindly carry a politician’s ideology or actions, while ‘bheda’ refers to those who follow their leaders or the crowd without much knowledge.

Our analysis shows that in the days leading up to the election, there was a high volume of positive comments, particularly targeting Shah, and negative comments targeting Keshav Sthapit and Sirjana Singh, which may have contributed to the polarity for them. Unfortunately, Facebook posts about election rallies and meetings by the latter two major political candidates were either sparsely commented on or met with negative sentiments.