<p>Paragraphs in document refer to various themes depending on the domain. Thematic classification is a task in NLP that aims to classify segments of text in a document based on its context. This research work focuses on documents from the legal domain containing paragraphs related to themes such as facts, arguments, and verdicts, which are referred to as rhetorical roles. These documents are complex, verbose, and involve domain-specific vocabulary, making it challenging to analyze. Paragraphs are strategically constructed to include information specific to a rhetorical role. Legal practitioners often need to analyze preceding relatable documents in order to access theme-wise information for tasks like summarization, similarity analysis, etc. This requirement raises the need to classify paragraphs into rhetorical roles. In this paper, a novel hybrid approach based on layout features such as paragraph position and sequence is proposed, along with semantic similarity of cue phrases for thematic classification of paragraphs. Results obtained outperform existing state-of-the-art techniques.</p>