Thesis
Using statistical models and sentiment analyses to better understand user engagement with YouTube music videos
- Creator
- Rights statement
- Awarding institution
- University of Strathclyde
- Date of award
- 2025
- Thesis identifier
- T17485
- Person Identifier (Local)
- 201850914
- Qualification Level
- Qualification Name
- Department, School or Faculty
- Abstract
- The increasing prevalence of social media platforms has transformed the way users consume, interact with, and share content. Platforms such as YouTube and Twitter serve as key media for music video dissemination, fostering dynamic user engagement and shaping audience sentiment. Despite their widespread use, limited research has examined the nuanced differences in user behaviour, emotional responses, and engagement patterns across these platforms. Addressing this gap, this thesis investigates user engagement, sentiment dynamics, and content strategies related to music videos across two major platforms, YouTube and Twitter. This research explores how textual, numerical, and time-related features, as well as user sentiment and platform-specific dynamics, influence engagement on YouTube and Twitter. Using machine learning models, alongside sentiment analysis techniques, the study evaluates the effectiveness of statistical models in predicting engagement, examines feature impacts, and assesses sentiment consistency and heterogeneity across platforms. Specifically, a mixed-method approach, combining machine learning and sentiment analysis, was used to analyze two datasets: 1,538 YouTube music videos and 76,171 comments from Twitter hyperlinks, and 2,119 YouTube music videos and 40,754 comments from YouTube channels. Feature combination strategies were tested to identify key predictors of user engagement, with model performance varying across datasets and feature configurations. While BDTR (Bagged Decision Tree Regression) demonstrated strong and consistent performance in several settings, it did not consistently outperform other models such as Gradient Boosting or Random Forest. Sentiment analysis conducted using the VADER lexicon-based approach, along with topic modelling via Latent Dirichlet Allocation (LDA), provided further insights into the emotional and topical dynamics of user discussions. By integrating these analyses, the research provides insights into platform-specific user behaviour and strategies to enhance content engagement. The findings highlight significant differences in how users engage with content creators on each platform. On YouTube, channel branding and creator influence play a crucial role in shaping user engagement, while Twitter interactions are more influenced by the emotional tone and topicality of content. Positive videos on both platforms rely on high quality content to drive engagement, whereas negative videos generate broader discussions and higher interaction rates due to their controversial nature. Additionally, while YouTube exhibits consistent emotional engagement across video categories, Twitter demonstrates greater emotional variability, reflecting the platform’s fast-paced and interactive nature. This research contributes to the growing body of knowledge on cross-platform user behaviour and engagement by integrating sentiment analysis, topic modelling, and predictive modelling. Specifically, it offers insights into how user interactions with music videos differ between YouTube and Twitter, shedding light on platform-specific engagement drivers such as emotional tone, topicality, and creator influence. These findings provide a framework for content creators, marketers, and platform managers to develop more effective strategies tailored to each platform’s unique dynamics. For music video audiences, this study helps clarify how engagement behaviours shape content visibility and influence which videos gain traction, potentially informing strategies that align with user interests and viewing patterns. However, this research has limitations. The focus on music video content may limit the generalizability of findings to other domains such as news or education. Additionally, the reliance on lexicon-based sentiment analysis introduces potential biases in interpreting emotional tones. Future research should explore a broader range of content types, examine emerging platforms such as TikTok or Instagram, and incorporate advanced sentiment analysis techniques, such as transformer-based models. In conclusion, this thesis enhances the understanding of user engagement dynamics and cross-platform content diffusion across YouTube and Twitter. By providing a comparative analysis of cross platform dynamics, it advances knowledge on digital media consumption and informs content creators about audience interaction patterns in the evolving social media landscape.
- Advisor / supervisor
- Chowdhury, G. G. (Gobinda G.)
- Revie, Crawford
- Resource Type
- DOI
Relations
Items
| Thumbnail | Title | Date Uploaded | Visibility | Actions |
|---|---|---|---|---|
|
|
PDF of thesis T17485 | 2025-11-11 | Public | Download |