Beyond sentiment: an integrated network-centric approach to trustworthiness dynamics in digital conversations
Beyond sentiment: an integrated network-centric approach to trustworthiness dynamics in digital conversations
In online discourse, audiences frequently encounter emotionally charged content that has been selectively amplified through editorial choice, algorithmic curation or both, yet they lack accessible tools to assess whether such content reflects the broader networked conversation. This thesis addresses this audience-centred problem by developing and empirically evaluating a novel socio-technical framework that supports trust inference in online social networks, grounded in Media Framing Theory and the credibility literature.
The research begins by systematically reviewing 27 computational trust frameworks using PRISMA guidance, confirming a persistent methodological divide between structural and content-based analyses. An initial exploratory experiment on a large COVID-19 Twitter dataset provides empirical evidence for this disconnect, revealing an "anger-periphery mismatch" where the most emotionally intense users are often structurally isolated. In response, this thesis proposes the SenCen framework, which operationalises affective trustworthiness by multiplicatively fusing a user's composite network centrality with their aggregated sentiment score. This approach yields a single, signed value that qualifies a user's structural reach with the affective character of their content.
The framework is then evaluated on a COVID-19 corpus through three complementary forms of analysis: a comparison of multiplicative and additive formulations, a sensitivity analysis across parameter variations, and a manual validation exercise on a stratified sample of users that achieved 90% acceptable agreement with human classification. The results demonstrate that the framework consistently identifies globally recognised public health authorities (such as the WHO and CDC) as the most affectively-trustworthy voices, while distinguishing them from structurally prominent voices with predominantly critical affective profiles.
The primary contribution of this thesis is a validated, context-aware analytical framework that supports trust inference without asserting trust as a direct measurement. By demonstrating that affective trustworthiness depends on the combination of structural reach and constructive affective tone, the SenCen framework offers a methodologically rigorous tool for audiences, researchers, and platform analysts to interpret the complex dynamics of online discourse during crisis communication.
Trust inference, Sentiment analysis, network centrality, Trustworthiness, Media framing, Crisis communication, Computational frameworks, Audience interpretation, Credibility measure, twitter analysis, Digital conversations, Social media analysis
University of Southampton
Alzaidi, Areej Hamad
7433059c-083d-4f6a-b253-e5c5ab553c94
2026
Alzaidi, Areej Hamad
7433059c-083d-4f6a-b253-e5c5ab553c94
Hall, Dame Wendy
11f7f8db-854c-4481-b1ae-721a51d8790c
Carr, Leslie
0572b10e-039d-46c6-bf05-57cce71d3936
Alzaidi, Areej Hamad
(2026)
Beyond sentiment: an integrated network-centric approach to trustworthiness dynamics in digital conversations.
University of Southampton, Doctoral Thesis, 148pp.
Record type:
Thesis
(Doctoral)
Abstract
In online discourse, audiences frequently encounter emotionally charged content that has been selectively amplified through editorial choice, algorithmic curation or both, yet they lack accessible tools to assess whether such content reflects the broader networked conversation. This thesis addresses this audience-centred problem by developing and empirically evaluating a novel socio-technical framework that supports trust inference in online social networks, grounded in Media Framing Theory and the credibility literature.
The research begins by systematically reviewing 27 computational trust frameworks using PRISMA guidance, confirming a persistent methodological divide between structural and content-based analyses. An initial exploratory experiment on a large COVID-19 Twitter dataset provides empirical evidence for this disconnect, revealing an "anger-periphery mismatch" where the most emotionally intense users are often structurally isolated. In response, this thesis proposes the SenCen framework, which operationalises affective trustworthiness by multiplicatively fusing a user's composite network centrality with their aggregated sentiment score. This approach yields a single, signed value that qualifies a user's structural reach with the affective character of their content.
The framework is then evaluated on a COVID-19 corpus through three complementary forms of analysis: a comparison of multiplicative and additive formulations, a sensitivity analysis across parameter variations, and a manual validation exercise on a stratified sample of users that achieved 90% acceptable agreement with human classification. The results demonstrate that the framework consistently identifies globally recognised public health authorities (such as the WHO and CDC) as the most affectively-trustworthy voices, while distinguishing them from structurally prominent voices with predominantly critical affective profiles.
The primary contribution of this thesis is a validated, context-aware analytical framework that supports trust inference without asserting trust as a direct measurement. By demonstrating that affective trustworthiness depends on the combination of structural reach and constructive affective tone, the SenCen framework offers a methodologically rigorous tool for audiences, researchers, and platform analysts to interpret the complex dynamics of online discourse during crisis communication.
Text
Areej_AlZaidi_Thesis_Ver3
- Version of Record
Text
Final-thesis-submission-Examination-Mrs-Areej-Alzaidi
Restricted to Repository staff only
More information
Published date: 2026
Keywords:
Trust inference, Sentiment analysis, network centrality, Trustworthiness, Media framing, Crisis communication, Computational frameworks, Audience interpretation, Credibility measure, twitter analysis, Digital conversations, Social media analysis
Identifiers
Local EPrints ID: 513684
URI: https://eprints.soton.ac.uk/id/eprint/513684
PURE UUID: 98123131-be00-4ec5-ab95-545b44e6dad2
Catalogue record
Date deposited: 24 Aug 2026 16:45
Last modified: 13 Sep 2026 15:50
Export record
Altmetrics
Contributors
Author:
Areej Hamad Alzaidi
Download statistics
Downloads from ePrints over the past year. Other digital versions may also be available to download e.g. from the publisher's website.
View more statistics