Thesis
Social media information credibility in the context of dementia
- Creator
- Rights statement
- Awarding institution
- University of Strathclyde
- Date of award
- 2022
- Thesis identifier
- T16413
- Person Identifier (Local)
- 201790662
- Qualification Level
- Qualification Name
- Department, School or Faculty
- Abstract
- Credibility is a major concern of social information retrieval (SIR). Absence of editorial oversight and increased bot presence on social media (SM) have led to the spread of low quality health information, which could be misinterpreted by vulnerable populations (e.g., people with dementia) and their caregivers, affecting their lives negatively. Several studies have proposed automated solutions to evaluate information quality on SM, yet most ignore the role of bots. Another limitation of previous research is that there is a lack of understanding of the features that affect user perception and automation solutions. Automation solutions also often rely on human annotation. Also, previous studies have mainly focused on social events and political topics, with limited focus on the health context. Therefore, the purpose of this research is to explore the credibility aspects, information quality and perceived credibility by humans, in a health context, focusing on dementia information on Twitter. This research employed a sequential explanatory mixed methods design and conducted three empirical studies in two phases. In the first phase, the research explored bot features in the context of dementia to evaluate the feasibility of using these features to automatically assess the quality of dementia information. Then, different annotated dementia related myth datasets were used to examine the usefulness of varying combinations of features developed in the first study and gleaned from the literature in assessing information quality using a quantitative approach with several supervised machine learning (ML) algorithms. The compiled classification ML model reached an accuracy score of 84% using 28 different linguistic and domain features. These promising results indicate that using the identified features in automatic assessment is feasible. In the second phase, a qualitative approach (using the think-aloud method) was used to identify the most crucial features from user perspectives by analysing people’s assessment of information credibility on SM when they were provided access to all the available features on the platform. The findings demonstrate the importance of the qualitative approach to expand understanding of perceived credibility from the user perspective. Users reported unique credibility factors associated with the particular context, and some of these factors are explained using Sundar’s MAIN model (Sundar, 2008). Employing this mixed methods design provided a holistic picture of the research problem. The research findings provide insight into the dissimilarity between information quality evaluated by automated methods and perceived credibility of information evaluated by information consumers. Evaluation by automated methods appears to be based mainly on static features (linguistic cues), whereas user evaluation reveals combinations of static and dynamic features influenced by consumer related characteristics, like prior knowledge and relevance of the information to the consumer. The outcome of this research points to the need for future research to close the gap between human and machine interpretation of credibility. This research concludes by proposing a framework that includes both features evaluated using ML and features based on consumer perception. The framework can be used to develop an automatic assessment model of health information credibility on SM.
- Advisor / supervisor
- Ayouni, Sara
- Pennington, Diane Rasmussen
- Resource Type
- Note
- THIS THESIS WAS PREVIOUSLY HELD UNDER MORATORIUM FROM OCTOBER 24 2022 UNTIL OCTOBER 24 2023.
- DOI
- Embargo Note
Relations
Items
| Thumbnail | Title | Date Uploaded | Visibility | Actions |
|---|---|---|---|---|
|
|
PDF of thesis T16413 | 2023-03-14 | Public | Download |