AutoDocDB: A Novel Framework for Automated Source Code Summarization in Database Management Systems via Advanced Text Extraction and CodeT5+ Integration

This study aims to develop a comprehensive framework, AutoDocDB, for automated source code summarization in database management systems (DBMS).The four primary objectives are: (1) to propose an effective method for text extraction from DBMS source code; (2) to study the feasibility of a pre-processing framework for the extracted text; (3) to investigate multi-faceted feature extraction techniques for representing the text; and (4) to introduce a novel technique for automatically generating summaries using CodeT5+.We curated a dataset of DBMS source code from textbooks and online repositories.Text was extracted using a hybrid parser that isolates code comments, string literals, and SQL queries.The extracted text was preprocessed using tokenization, stop-word removal, and syntax-aware cleaning.Features were extracted using a combination of TF-IDF, BERT embeddings, and AST-based structural features.The CodeT5+ model was fine-tuned on the processed data to generate summaries.The hybrid text extraction method achieved 95\% accuracy in identifying relevant text components.The pre-processing framework improved feature quality by 30\%.CodeT5+ fine-tuned with extracted features achieved a BLEU score of 72\% and a ROUGE-L score of 75\%, outperforming baseline models.AutoDocDB effectively addresses all four objectives, providing a robust pipeline for DBMS code summarization.The integration of advanced text extraction, pre-processing, feature engineering, and CodeT5+ fine-tuning sets a new standard for automated documentation in domainspecific languages.IndexTerms -source code summarization;