<p>Semantic legal data offers the basis for a methodical examination of legal provisions and is vital for comprehending and deciphering legal regulations. Nevertheless, manually adding semantic metadata to sizable criminal datasets is expensive and time-consuming. The cutting-edge study addresses two essential troubles: the requirements engineering (RE) literature lacks a standardized framework for semantic metadata types relevant to prison requirements evaluation, and there may be insufficient automatic guide for extracting those metadata types, especially while using deep learning (DL) and Natural language processing (NLP) capabilities. To address those problems, a comprehensive framework was first created via reviewing and integrating the semantic criminal metadata categories determined inside the RE literature. After that, an automated extraction technique that makes use of NLP was created to help realize and examine legal texts, enabling to find and categorize critical metadata. A Binary Moth-Flame Optimized Dynamic recurrent neural Network (BMFO-DRNN) is then used to broaden an automatic extraction approach for the specified metadata lessons. To increase the satisfactory and relevance of the entered statistics, preprocessing techniques, which include tokenization, stemming, lemmatization, and word embeddings like Word2Vec, were used for characteristic extraction. Experimental result shows the BMFO-DRNN model outperforms traditional methods, achieving accuracy (95%), F1-score (92%), precision (93%), and recall (96%). In addition to demonstrating the price of NLP in automated semantic analysis legal metadata, this work additionally shows how NLP should improve the efficacy and performance of legal evaluation in an effort to assist criminal informatics advancement.</p>