Visual Methods for Network Analytics of Echo Chamber: A Case Study

of Thailand’s General Election 2023

Isariyaporn Sukcharoenchaikul and Puripant Ruchikachorn

Chulalongkorn University, Thailand

Keywords:

Echo Chamber, Visualization, Visualization Techniques, Network Analytics, Network Graph Analysis,

Political Discussions, Social Media Analysis.

Abstract:

This research develops visual methods to study the echo chamber effect through a case study on Thai-

land’s 2023 General Election. Using visualization techniques like node-link diagrams, t-SNE projections,

and heatmaps, it examines homophilic relationships, clustering, and polarization in online communities. To

minimize inaccuracies and biases, network graphs are created from contextual analysis of user-generated con-

tent, rather than relying on predeﬁned relationships like friendships or followers. The study applies the Echo

Chamber Score (ECS) with visualizations to explore variations in analytical methods and how they capture

different aspects of echo chambers. Additionally, it illustrates how political events shape online discourse and

community dynamics by linking ECS with key political milestones.

1 INTRODUCTION

The rise of social media has intensiﬁed echo cham-

bers, where selective exposure reinforces pre-existing

beliefs, promoting polarization, misinformation, and

conspiracy theories. Platforms like Twitter, Face-

book, and Pantip amplify these effects, particularly

during events like Thailand’s 2023 General Election

(Vicario et al., 2016)(Unerman, 2020).

This study employs network-based analytical

methods to examine echo chamber formation. It con-

structs network graphs based on user-generated con-

tent, semantic similarity, and shared opinions, utiliz-

ing visualization techniques such as node-link dia-

grams, t-SNE projections, and heatmaps, alongside

the EchoGAE model (Alatawi et al., 2023).

By analyzing graph construction methodologies

and external inﬂuences, this research integrates com-

putational social science with visual analytics to pro-

vide insights into selective exposure and polarization.

The ﬁndings offer a structured approach to under-

standing echo chambers and their societal impact.

2 RELATED WORK

This section discusses key studies in network graph

analysis, homophilic interactions, and echo chambers,

focusing on methods used to study ideological polar-

ization in political discussions.

2.1 Echo Chamber

An echo chamber is an environment where selective

exposure reinforces beliefs, limiting diverse perspec-

tives and amplifying conﬁrmation bias (Cinelli et al.,

2021)(Jiang et al., 2021). Social media exacerbates

this effect by directing users to supportive content (Vi-

cario et al., 2016). Events like the 2016 USA elec-

tion demonstrated how echo chambers distort percep-

tions (Guo et al., 2018), though awareness of the is-

sue slightly reduced their effect by 2020 (Yang et al.,

2020)

Traditional approaches to quantifying echo cham-

bers rely on modularity analysis, random walkers,

and opinion-spreading models (Markgraf and Schoch,

2019)(Cota et al., 2019), but these often require ex-

tensive ideological labeling. EchoGAE, a Graph

Autoencoder-based model, overcomes this limita-

tion by generating embeddings that integrate content-

based similarity with network structure, allowing

for scalable analysis of polarization (Alatawi et al.,

2023). Another approach to studying echo cham-

bers is through signals like polarization, which leads

to ideologically segregated communities. These can

be detected using methods such as the Louvain algo-

rithm and visualized with density plots and clustering

374

Sukcharoenchaikul, I., Ruchikachorn and P.

Visual Methods for Network Analytics of Echo Chamber: A Case Study of Thailand’s General Election 2023.

DOI: 10.5220/0013508900003967

In Proceedings of the 14th International Conference on Data Science, Technology and Applications (DATA 2025), pages 374-381

ISBN: 978-989-758-758-0; ISSN: 2184-285X

techniques (Conover et al., 2011)(Botte et al., 2022),

which are often more interpretable compared to ma-

chine learning methods like EchoGAE. This study

builds on previous work by combining network anal-

ysis to interpret echo chamber signals and employing

EchoGAE to analyze clustering, homophilic interac-

tions, and ideological segregation

2.2 Network Graph Construction

Earlier social network research focused on prede-

ﬁned relationships (Goodreau et al., 2009), but recent

work emphasizes contextual interactions (Mcpherson

et al., 2001)(Goel et al., 2019). Textual analysis

methods like LDA and k-means clustering reveal pat-

terns in user interactions (Hoque and Carenini, 2014).

While deep learning models like Sentence-BERT of-

fer precise insights (Reimers and Gurevych, 2019),

this study balances efﬁciency and interpretability by

using LDA and k-means to explore thematic cluster-

ing.

2.3 Visualization for Echo Chamber

Analysis

Visualization reveals patterns in complex network

data. Node-link diagrams, force-directed layouts, and

hybrid techniques like Nodetrix (Henry et al., 2007)

highlight community structures. Visual analytics in-

tegrates computational methods with interactive visu-

alization (Thomas and Cook, 2005). Tools like So-

cialOcean facilitate interaction analysis(Diehl et al.,

2018), while t-SNE visualizations highlight ideolog-

ical clustering and polarization(van der Maaten and

Hinton, 2008). This study uses basic visualization

techniques to analyze network graphs, support echo

chamber exploration, and incorporate EchoGAE into

visual analytics to examine polarization and varia-

tions in graph construction.

3 DESIGN REQUIREMENT

ANALYSIS

This study designs visual methods for echo chamber

analysis. The visualizations should provide insights

based on the following requirements.

Network Graph Characteristics. The goal of this

visualization design is to explore the structure of net-

work graphs and address key questions, such as: What

does the network graph look like? Are there observ-

able patterns indicative of echo chamber effects? To

capture echo chambers effectively, network graphs

must reﬂect homophilic interactions—users interact-

ing with others who share similar attributes, such as

ideologies or opinions. This tendency, known as ho-

mophily, is central to echo chamber dynamics, as

users reinforce each other’s views and isolate dissent-

ing opinions (Mcpherson et al., 2001). Differences

in construction methods reveal unique characteristics,

aiding method selection.

Echo Chamber Effect. Building on the exploration

of echo chamber signals in network graphs, the next

question is: How do user attributes inﬂuence the abil-

ity to capture echo chambers? To answer this, the

EchoGAE embedding model is employed. Visualiz-

ing user embeddings enhances understanding of the

network while incorporating user attributes. The vi-

sualizations should enable researchers to explore how

network characteristics, such as homophily, and em-

bedded node attributes contribute to echo chamber

formation. Success will be measured by how clearly

the visualizations reveal clustering patterns and inter-

actions aligned with echo chamber phenomena.

News Inﬂuence on Echo Chamber Formation.

This research also examines how factors, particularly

news events, affect echo chamber formation. Al-

though algorithms are not considered here, it is as-

sumed that echo chambers are more likely to form

when discussions attract signiﬁcant attention, beyond

just topic categorization (e.g., controversial vs. non-

polarized topics) (Alatawi et al., 2023). This sec-

tion focuses on how key news events, varying in at-

tention and impact, inﬂuence echo chamber develop-

ment. The visualizations will demonstrate how news

events contribute to the formation and intensity of

echo chambers by tracking monthly events and quan-

tifying their effects.

4 VISUALIZATION

This section details the visual methods applied to net-

work structures, following the design framework.

4.1 Network Graph Structure Overview

Network graphs are powerful tools for studying the

structural patterns and dynamics of social interac-

tions, offering insights into user behaviors, commu-

nity formations, and the prevalence of homophilic

relationships. By visualizing these interactions, re-

searchers can uncover the underlying structure of so-

cial networks and examine how connections between

Visual Methods for Network Analytics of Echo Chamber: A Case Study of Thailand’s General Election 2023

375

individuals shape broader patterns of discourse and

group behavior.

Network Graph Structure. Node-link diagrams

reveal network patterns. In Figure 1, nodes repre-

sent users, and edges indicate homophilic interac-

tions. Densely connected clusters suggest stronger

echo chambers.

Community Structure This visualization is intro-

duced to examine community structure patterns. The

Louvain algorithm identiﬁes communities (Figure 3),

with nodes representing groups, edges showing inter-

community interactions and nodes colors indicate the

number of users in group. This highlights ideological

clustering and reveals how different methods capture

network structures.

Method Comparison. Comparing network graph

construction methods ensures accurate homophilic re-

lationship representation. A sequence of visualiza-

tions examines structural differences, accompanied

by a bar chart quantifying unique and common edges.

Figure 4.

4.2 Echo Chamber Effect View

For echo chamber exploration, the EchoGAE model

is utilized to embed user attributes into node rep-

resentations. A t-SNE plot is employed to project

high-dimensional user interaction data into a two-

dimensional space (Figure 5), facilitating the visual-

ization of clustering patterns associated with the echo

chamber effect. Each point in the plot represents a

user, with spatial proximity indicating similarity in in-

teractions or opinions. Users who cluster closely to-

gether are likely to share similar viewpoints or engage

frequently, potentially forming echo chambers.

This visualization is adapted from (Alatawi et al.,

2023), where the t-SNE method was applied to il-

lustrate clustering patterns after embedding node at-

tributes using EchoGAE. The results reveal cluster-

ing tendencies, where dense, isolated clusters suggest

strong echo chambers, while overlapping or loosely

connected clusters may indicate weaker echo cham-

bers or more diverse interactions.

4.3 News Inﬂuence on the Formation of

Echo Chambers

The visualizations discussed thus far focus on iden-

tifying echo chamber signals and analyzing network

graph characteristics to support method selection. In

contrast, this visualization examines factors related to

echo chambers using the Echo Chamber Score (ECS),

with a focus on news and events. A heatmap (Fig-

ure 6) is used to compare the engagement levels of key

events with the ECS over time, exploring the potential

impact of news events on echo chamber formation.

5 EVALUATION

In this section, the effectiveness and usability of the

visual methods are demonstrated through case stud-

ies, using Thailand’s 2023 General Election discus-

sion data from Pantip. This dataset was selected

as it provides real-world discourse on a widely dis-

cussed event. While political events often involve

diverse perspectives and varying degrees of polariza-

tion, the analysis in this study remains methodologi-

cally driven, focusing on the structural and behavioral

patterns within the data rather than making norma-

tive judgments. The use of real-world data highlights

the practical applicability of the proposed methods in

examining online discourse dynamics in an empirical

context.

5.1 Political Polarization and

Ideological Echo Chambers

The case study examines ideological polarization in

the Rajdumnern forum discussions during the elec-

tion. The dataset includes 10,771 posts and 150,487

comments, embedded using WangchanBERTa (Low-

phansirikul et al., 2021) to capture semantic meaning.

A network graph was constructed with nodes as users

and edges indicating agreement or alignment, repre-

senting homophilic interactions.

Two techniques, k-means and Latent Dirichlet Al-

location (LDA), were applied to group users based on

semantic similarity. Networks were then created from

two distinct relationship deﬁnitions, each designed to

capture homophilic interactions:

Semantic Agreement Connectivity(SAC): An

edge is established if a user’s comment belongs to the

same cluster as the original post, indicating direct se-

mantic agreement on political content. This approach

assumes that homophilic interactions arise from users

responding to posts.

Interactive Alignment Connectivity(IAC): Based

on web forum behavior, an edge is formed if two users

comment on the same post and their comments clus-

ter together. This treats comments as content nodes,

capturing indirect alignment among commenters.

By combining these two clustering techniques

with the two relationship deﬁnitions, four distinct net-

DATA 2025 - 14th International Conference on Data Science, Technology and Applications

376

work graphs were generated, each providing a unique

perspective on user alignment. This approach enables

a comprehensive comparison of how different clus-

tering methods and relationship deﬁnitions inﬂuence

network construction, offering nuanced insights into

ideological polarization.

5.1.1 Identifying Ideological Polarization

Through Network Structure

The node-link diagram in Figure 1 illustrates the

network structure, where nodes represent users and

edges indicate homophilic interactions based on po-

litical opinion alignment. A central cluster domi-

nates all graphs, reﬂecting a high concentration of

users with similar viewpoints and reinforcing the echo

chamber effect.

(a) k-means-SAC (b) k-means-IAC

Figure 1: Comparison of Network Graphs Using Four Dif-

ferent Methods.This ﬁgure presents network graphs con-

structed using two clustering methods (k-means and LDA)

and two relationship deﬁnitions (SAC and IAC). The colors

represent the network group to which each node belongs

based on its connections.

Graphs from SAC further highlight this trend,

showing users gravitating toward shared content,

which aligns with selective exposure issues reported

in prior research (Barber

a et al., 2015). Smaller, iso-

lated groups at the edges likely represent users with

differing opinions or less frequent interactions, while

the dense center indicates high engagement among

like-minded users.

To analyze ideological polarization, the Louvain

algorithm was applied for community detection, vi-

sualized in Figure 1. Given the complexity of large

networks, only the top 500 highest-degree nodes were

selected to improve readability and interpretability.

The Louvain algorithm revealed multiple commu-

(a) k-means-SAC (b) k-means-IAC

Figure 2: Network Community Structure Focusing on

High-Degree Nodes. This ﬁgure displays subgraphs con-

sisting of the top 500 high-degree nodes from the network.

The nodes are colored based on the community they belong

to, as detected using the Louvain algorithm.

nities within the assumed central cluster, weakening

its cohesion Figure 2. It is suggests a ﬂuid network

structure rather than strongly separated ideological

groups. High-centrality nodes act as bridges, indicat-

ing that echo chambers may form through informa-

tion ﬂow rather than strict divisions. These ﬁndings

challenge the assumption that a dense central cluster

represents an echo chamber, highlighting the role of

selective engagement and issue-based agreement in

shaping ideological divisions.

Real data analysis reinforces these ﬁndings, show-

ing that ideological divisions are complex and issue-

based rather than strictly partisan. While LDA-SAC

highlights community separation, engagement pat-

terns suggest political alignment is ﬂuid, with users

selectively agreeing on policies rather than adhering

strictly to party lines. This leads to overlapping com-

munities rather than entirely isolated ideological clus-

ters.

Overall, these visualization meets the key require-

ments and provides a comprehensive view of the data.

However, it lacks clear differentiation when compar-

ing the various methods. Additionally, handling large

datasets presents a challenge, and the visualization

could be further enhanced to address this limitation,

ultimately improving its effectiveness and scalabil-

ity(von Landesberger et al., 2011).

To conﬁrm ideological polarization and explore

signs of the echo chamber effect, the Community

Structure visualization was applied to visualize con-

nections between communities, and the number of

users within each community (Figure 3).

Visual Methods for Network Analytics of Echo Chamber: A Case Study of Thailand’s General Election 2023

377

(a) k-means-SAC (b) k-means-IAC

Figure 3: Visualization of Community Relationship in

Network Graphs. This ﬁgure presents an interactive net-

work graph, showing cluster relationships for each con-

struction method. Nodes represent communities identiﬁed

by the Louvain algorithm, with colors indicating commu-

nity size. Edges represent homophilic interactions between

users from different communities.

The colors in the visualization represent cluster

sizes, with a gradient from purple for the largest clus-

ters to green and yellow for smaller ones. Nodes col-

ored in purple indicate the largest, most central clus-

ters, suggesting they represent the most active groups

in the network.

Across methods, k-means-SAC (Figure 3a) shows

a hierarchical structure with isolated groups, LDA-

SAC (Figure 3c) reveals stronger interconnections, k-

means-IAC (Figure 3b) highlights denser core struc-

tures, and LDA-IAC (Figure 3d) balances central den-

sity and decentralization. Sparse inter-group links

across all methods indicate polarization-driven echo

chambers.

Network structures reﬂect issue-based homophily

rather than strict political divisions. LDA-SAC, cap-

turing ideological clustering through semantic agree-

ment, shows that echo chambers emerge from the-

matic alignment rather than absolute polarization.

5.1.2 Comparing Structural Patterns Across

Methods

Graphs were plotted to compare relationships across

the four methods (Figure 4), with edge colors dis-

tinguishing unique and shared relationships across

11 characteristics. The bar chart illustrates the dis-

tribution of unique and common edges across G1

(k-means-SAC), G2 (LDA-SAC), G3 (k-means-IAC),

(a) Comparison of Relationship across Four Methods

(b) Edges Counts: Unique and Common Edges Across

Graphs

Figure 4: Comparison of Relationships Across Four Meth-

ods This ﬁgure compares relationships in four network

graphs constructed using k-means and LDA clustering un-

der SAC and IAC. The node-link diagram (top) shows the

network structure, and the bar chart (bottom) quantiﬁes the

distribution of unique and common edges.

and G4 (LDA-IAC), revealing structural complexity.

The graph in Figure 4 highlights unique relation-

ships across the four methods, revealing structural

complexity. The bar chart displays the number of

edges unique to each method (k-means-SAC, LDA-

SAC, k-means-IAC, LDA-IAC) and shared edges, of-

fering a clear breakdown of edge distributions.

LDA-IAC captures the most nuanced relation-

ships, as shown by its higher number of unique edges

(Figure 4b). The high common-edge count in IAC-

based methods (G3, G4) highlights their sensitivity to

subtle homophilic connections.

Selecting LDA-SAC balances content-driven co-

herence with network clarity, effectively capturing

semantic-based homophily while maintaining inter-

pretability in network structure.

DATA 2025 - 14th International Conference on Data Science, Technology and Applications

378

5.1.3 Exploring Ideological Alignment Using

User Embeddings

The Echo Chamber Effect is measured using the

EchoGAE model, which requires the network graph

(see Section 5.1) and user embeddings, created by av-

eraging 20 posts or comments per user with normal-

ized missing data. These inputs allow the model to

assess interactions, with ECS derived from the silhou-

ette score to measure cohesion and separation.

(a) k-means-SAC (b) k-means-IAC

Figure 5: 2D Projections of User Embeddings Across Four

Network Graphs. This ﬁgure presents the 2D projections

(using the t-SNE algorithm (van der Maaten and Hinton,

2008)) of user embeddings from four different network

graphs after embedding using EchoGAE algorithm. Col-

ors represent distinct communities.

The analysis of political conversations during

Thailand’s 2023 general election reveals that SAC-

based methods yield lower ECS, with LDA (0.240)

forming more homogeneous communities and k-

means (0.213) capturing diverse groups. IAC-based

methods produce higher ECS, particularly k-means

(0.415), indicating stronger polarization, while LDA

(0.315) shows moderate polarization. These results

highlight how clustering methods shape the echo

chamber effect.

The t-SNE visualization (Figure 5a, Figure 5c)

shows ideological alignment through a curve-like pro-

jection, indicating shared opinions drive user cluster-

ing, but ideological alignment is localized around spe-

ciﬁc topics.

Despite its lower ECS, LDA-SAC was selected for

its interpretable structure, capturing issue-based ho-

mophily. This choice aligns with prior polarization

ﬁndings but reveals discrepancies, suggesting that ex-

cluding node attributes may have contributed to the

inconsistency, requiring further exploration.

5.2 Temporal Analysis of Echo

Chamber Dynamics During Key

Election Events

Data on key events from January to August 2023 was

collected from news articles and social media, focus-

ing on Thailand’s general election. Engagement rates

were estimated using YouTube view counts from 30

relevant videos, retrieved via the YouTube API with

keywords such as ”Thailand Election 2023”, ”Thai-

land General Election”.

Engagement levels—Very High, High, Moder-

ate, and Low—were based on log-transformed view

counts, with quartiles deﬁning the categories. These

classiﬁcations reﬂect public interest and media atten-

tion, with major events like the election labeled as

Very High engagement, and routine updates as Mod-

erate or Low.

Figure 6: Heatmap of ECS by Month, Showing Engage-

ment Rate Levels Estimated Based on the Importance of

Key Events in Thailand’s 2023 General Election.

The heatmap (Figure 6) shows ECS by month

from the LDA-SAC method, alongside engagement

rate levels, which are set as a proportion of the max-

imum ECS. Four engagement levels—Low, Moder-

ate, High, and Very High—correspond to 25%, 50%,

75%, and 100% of the maximum ECS value.

Despite high engagement during key political

events, ECS does not consistently align with engage-

ment levels, suggesting that high engagement does

not always lead to stronger ideological polarization.

Major events may involve diverse discussions, dilut-

ing ideological boundaries. For example, the higher

ECS in June and July, despite varying engagement,

indicates that speciﬁc issues drove stronger align-

Visual Methods for Network Analytics of Echo Chamber: A Case Study of Thailand’s General Election 2023

379

ment, even though they were not discussed uniformly

throughout the year.

This highlights how LDA-SAC captures cluster-

ing around polarized issues, which may not always

correlate with engagement levels. The discrepancy

between ECS and engagement suggests that polariza-

tion is driven more by discourse focus than by in-

teraction volume. In some cases, lower engagement

with focused discussions leads to stronger alignment,

while high engagement with diverse topics weakens

clustering.

LDA-SAC reﬂects issue-based homophily, despite

ﬂuctuating engagement, justiﬁes its use for analyzing

political discourse during Thailand’s 2023 election.

Its focus on issue-driven discussions makes it a valu-

able tool for understanding ideological polarization in

political events.

6 DISCUSSIONS AND FUTURE

WORK

This research demonstrates the value of visual meth-

ods in examining echo chambers in online political

discussions, though several limitations exist. The use

of predeﬁned clustering techniques and network con-

struction methods may inﬂuence the detected struc-

tures, potentially missing alternative patterns of ideo-

logical alignment.

Methodological Implications and Limitations.

There are potential biases in network construction

and clustering parameters. The reliance on Pantip

data limits the generalizability of the ﬁndings,

highlighting the need for future research to validate

these methods across diverse platforms. Scalability

and interpretability remain key challenges, as large

networks can lead to visual clutter, and community

detection outcomes may vary depending on algorithm

selection. Future studies should incorporate weighted

edges to account for interaction strength and explore

methodologies to enhance both scalability and

interpretability.

This study does not make normative judgments on

political alignment but employs visual methods to an-

alyze structural patterns. Future research should re-

ﬁne the understanding of echo chamber dynamics by

examining peripheral communities and the evolution

of online discourse.

Expanding Visual Analysis Techniques. Future

work could incorporate advanced dimensionality re-

duction techniques, such as UMAP, for better user

clustering and high-dimensional structure preserva-

tion. Dynamic visualizations tracking changes in

echo chamber intensity over time would enable re-

searchers to observe how polarization evolves during

major political events or shifts in user interactions.

Exploring Platform Inﬂuence and User Behavior.

Echo chambers tend to intensify during political

events due to increased engagement. Future studies

should examine the inﬂuence of platform algorithms,

such as recommendation systems, on exposure to di-

verse viewpoints. Testing algorithm changes could

help reduce polarization while maintaining user en-

gagement.

Enhancing Echo Chamber Metrics and Applica-

tions. While the visual methods provided valuable

insights, developing new metrics could capture sub-

tler aspects of echo chambers, such as exposure to

contrasting opinions or content-based polarization.

These metrics could be used in cross-platform stud-

ies or to evaluate interventions aimed at reducing po-

larization.These metrics could be applied in cross-

platform studies or used to evaluate interventions

aimed at reducing polarization. Real-time social me-

dia monitoring could also help identify emerging echo

chambers and enable timely interventions to promote

balanced discourse.

Future work should incorporate quantitative met-

rics, such as clustering quality, modularity, and corre-

lation with expert classiﬁcations, to objectively assess

how different methods capture echo chamber phe-

nomena. A comparative table or performance evalua-

tion would enhance transparency and provide clearer

guidance on method selection.

7 CONCLUSION

This study developed visual methods for Network An-

alytics of Echo Chambers, with a speciﬁc case study

on Thailand’s General Election 2023. By constructing

network graphs using various clustering techniques

and relationship deﬁnitions, the research explored the

formation and dynamics of echo chambers in politi-

cal conversations. The ECS was employed to quantify

polarization within online communities, revealing the

inﬂuence of both method choice and political events

on user behavior and discourse.

The ﬁndings highlighted the varying levels of po-

larization across different methods, particularly not-

ing the stronger echo chamber effects observed un-

der the k-means clustering method. Additionally, the

research illustrated how signiﬁcant political events,

DATA 2025 - 14th International Conference on Data Science, Technology and Applications

380

such as news cycles and electoral milestones, played

a key role in amplifying echo chamber dynamics.

Overall, the visual methods developed in this

study offer visual methods for understanding the com-

plexities of online political discussions and the for-

mation of echo chambers. The case study provides

important insights into the impact of the 2023 gen-

eral election in Thailand on online community dy-

namics. This work lays the foundation for future re-

search and applications in visualizing and analyzing

network-based social phenomena in the context of po-

litical communication.

REFERENCES

Alatawi, F., Sheth, P., and Liu, H. (2023). Quantifying the

echo chamber effect: An embedding distance-based

approach.

Barber

a, P., Jost, J. T., Nagler, J., Tucker, J. A., and Bon-

neau, R. (2015). Tweeting from left to right. Psycho-

logical Science, 26:1531 – 1542.

Botte, N., Ryckebusch, J., and Rocha, L. E. (2022). Cluster-

ing and stubbornness regulate the formation of echo

chambers in personalised opinion dynamics. Phys-

ica A: Statistical Mechanics and its Applications,

599:127423.

Cinelli, M., Morales, G. D. F., Galeazzi, A., Quattrocioc-

chi, W., and Starnini, M. (2021). The echo chamber

effect on social media. Proceedings of the National

Academy of Sciences, 118(9):e2023301118.

Conover, M., Ratkiewicz, J., Francisco, M., Gonc¸alves, B.,

Menczer, F., and Flammini, A. (2011). Political polar-

ization on twitter.

Cota, W., Ferreira, S., Pastor-Satorras, R., and Starnini, M.

(2019). Quantifying echo chamber effects in infor-

mation spreading over political communication net-

works.

Diehl, A., Hundt, M., Haussler, J., Seebacher, D., Chen,

S., Cilasun, N., Keim, D., and Shreck, T. (2018). So-

cialocean: Visual analysis and characterization of so-

cial media bubbles. pages 1–11.

Goel, A., Ma, K. T., and Tan, C. (2019). An end-to-end net-

work for generating social relationship graphs. 2019

IEEE/CVF Conference on Computer Vision and Pat-

tern Recognition (CVPR), pages 11178–11187.

Goodreau, S. M., Kitts, J. A., and Morris, M. M. (2009).

Birds of a feather, or friend of a friend? using expo-

nential random graph models to investigate adolescent

social networks*. Demography, 46:103–125.

Guo, L., Rohde, J. A., and Wu, H. D. (2018). Who is re-

sponsible for twitter’s echo chamber problem? evi-

dence from 2016 u.s. election networks. Information,

Communication & Society, 23:234 – 251.

Henry, N., Fekete, J.-D., and McGufﬁn, M. J. (2007). Node-

trix: a hybrid visualization of social networks. IEEE

Transactions on Visualization and Computer Graph-

ics, 13(6):1302–1309.

Hoque, E. and Carenini, G. (2014). Convis: A visual

text analytic system for exploring blog conversations.

Computer Graphics Forum, 33(3):221–230.

Jiang, B., Karami, M., Cheng, L., Black, T., and Liu, H.

(2021). Mechanisms and attributes of echo chambers

in social media.

Lowphansirikul, L., Polpanumas, C., Jantrakulchai, N.,

and Nutanong, S. (2021). Wangchanberta: Pretrain-

ing transformer-based thai language models. ArXiv,

abs/2101.09635.

Markgraf, M. and Schoch, M. (2019). Quantiﬁcation of

echo chambers: a methodological framework consid-

ering multi-party systems. In European Conference

on Information Systems.

Mcpherson, M., Smith-Lovin, L., and Cook, J. (2001).

Birds of a feather: Homophily in social networks. An-

nual Review of Sociology, 27:415–.

Reimers, N. and Gurevych, I. (2019). Sentence-bert: Sen-

tence embeddings using siamese bert-networks. In

Conference on Empirical Methods in Natural Lan-

guage Processing.

Thomas, J. J. and Cook, K. A. (2005). Illuminating the

path: The research and development agenda for visual

analytics.

Unerman, J. (2020). Risks from self-referential peer review

echo chambers developing in research ﬁelds: 2018

keynote address presented at the british accounting re-

view 50th anniversary celebrations, british account-

ing and ﬁnance association annual conference, lon-

don. The British Accounting Review, 52(5):100910.

van der Maaten, L. and Hinton, G. (2008). Viualizing data

using t-sne. Journal of Machine Learning Research,

9:2579–2605.

Vicario, M. D., Bessi, A., Zollo, F., Petroni, F., Scala,

A., Caldarelli, G., Stanley, H. E., and Quattrocioc-

chi, W. (2016). The spreading of misinformation on-

line. Proceedings of the National Academy of Sci-

ences, 113(3):554–559.

von Landesberger, T., Kuijper, A., Schreck, T., Kohlham-

mer, J., van Wijk, J., Fekete, J.-D., and Fellner, D.

(2011). Visual analysis of large graphs: State-of-the-

art and future research challenges. Computer Graph-

ics Forum, 30(6):1719–1749.

Yang, T., Maj

o-V

azquez, S., Nielsen, R. K., and Gonz

alez-

Bail

on, S. (2020). Exposure to news grows less

fragmented with an increase in mobile access. Pro-

ceedings of the National Academy of Sciences,

117(46):28678–28683.

Visual Methods for Network Analytics of Echo Chamber: A Case Study of Thailand’s General Election 2023

381