<p>Scientific progress relies on a complex and interconnected network of scholarly publications housed within digital libraries. Although citations serve as the primary mechanism for linking this knowledge, a reference alone does not capture the rich contextual information in which the cited work is discussed. This limitation poses challenges for digital libraries attempting to accurately analyze scholarly influence. To address this issue, we use Citation Context Extraction (CCE) that is a foundational task for transforming raw citation links into meaningful, semantically enriched representations that can support advanced bibliometric and knowledge graph analyses. In this paper, we propose a novel twofold methodology to enhance CCE. First, we introduce an improved evidence-based extraction framework that leverages a richer set of linguistic and statistical signals to more accurately identify citation context sentences, extending beyond similarity-driven state-of-the-art approaches. Second, we propose an LLM-based framework that employs structured prompt engineering to enable deeper, more nuanced, and more explainable semantic interpretation of extracted citation contexts. We evaluate our methods on two distinct corpora: ACL-ARC, a domain-specific dataset in computational linguistics, and SDP-ACT, a multidisciplinary dataset spanning multiple scientific fields. Comparative experiments against established baselines demonstrate consistent improvements in citation context quality. Our findings contribute toward the development of more intelligent, interpretable, and semantically grounded digital library systems capable of mapping scholarly discourse and intellectual lineage more effectively.</p>