Author ORCID Identifier:
Date of Graduation
7-2026
Document Type
Dissertation
Degree Name
Doctor of Philosophy in Computer Science (PhD)
Degree Level
Graduate
Department
Computer Science & Computer Engineering
Advisor/Mentor
Gauch, Susan
Committee Member
Zhang, Lu
Second Committee Member
Yang, Song
Third Committee Member
Pan, Yanjun
Keywords
Arificial Intelligence; Generative AI; Misinformation Detection; Natural Language Processing; Retrieval Augmented Generation; Social Media Analytics
Abstract
Social media platforms have become central to information exchange, shaping public opinion across social, political, and economic domains. However, the massive volume of user-generated content, combined with its informal, nuanced, and often noisy nature, presents significant challenges for automated analysis and generation. Tasks such as stance detection, rumor verification, and personalized content generation are further complicated by sarcasm, evolving discourse structures, and diverse user preferences. Addressing these challenges requires models that can effectively leverage linguistic nuance, conversational dynamics, and collaborative user signals. Transfer learning has emerged as a powerful paradigm for improving performance in low-resource and complex language understanding tasks. However, its effectiveness depends heavily on the relationship between source and target tasks. To address limitations in stance detection, this work investigates sarcasm detection as an intermediate task, enabling models to capture subtle linguistic cues that influence stance interpretation. By integrating transformer-based architectures with convolutional and sequential layers, the proposed approach improves classification performance and generalizes effectively across both in-domain and cross-target settings, mitigating the scarcity of labeled data for emerging topics. While improved stance detection enhances text-level understanding, rumor verification requires deeper modeling of conversational context and discourse evolution. Existing approaches often rely on flat text representations, failing to capture how interactions unfold within discussion threads. To address this limitation, this work introduces a series of discourse-aware models that incorporate stance-conditioned representations, hierarchical structures, and temporal dynamics. By combining source and reply embeddings, encoding stance progression, and leveraging structural traversal strategies, these models capture both semantic content and interaction patterns. Attention-based aggregation and stance-aware structural features further enable scalable and interpretable representations, leading to improved accuracy, robustness, and early detection of misinformation across multiple benchmark datasets. Beyond analysis, personalized generation systems must retrieve and utilize relevant information tailored to individual users. Traditional retrieval-augmented generation approaches focus primarily on individual user profiles, often overlooking the value of shared preferences across similar users. To address this gap, this work proposes ClusterRAG, a collaborative framework that organizes users into semantically coherent clusters and performs retrieval at both cluster and document levels. By jointly leveraging signals from the target user and similar users, ClusterRAG enhances retrieval quality while maintaining efficiency and compatibility with diverse retrieval and generation models. In all these endeavors, extensive experiments are conducted on benchmark datasets to evaluate the effectiveness of the proposed methods and compare them with state-of-the-art approaches. The results demonstrate consistent improvements in Macro-F1 scores, accuracy, and generalization across stance detection and rumor verification tasks, as well as enhanced performance in personalized retrieval-augmented generation. Additional analyses highlight the contributions of transfer learning, discourse-aware modeling, and collaborative filtering in improving robustness and scalability. Together, these contributions establish a comprehensive framework for advancing social media analytics and personalized generation. By integrating transfer learning, discourse-aware representations, and collaborative modeling, this dissertation provides practical and scalable solutions for understanding and generating user-centered content in complex social media environments.
Citation
Nkhata, G. (2026). Advancing Social Media Analytics and Personalized Generation via Transfer Learning, Discourse-Aware Modeling, and Collaborative Modeling. Graduate Theses and Dissertations Retrieved from https://scholarworks.uark.edu/etd/6372
