Knowledge Graph Creation based on Ontology from Source-Code: The Case of C#

The software's core data and business logic are believed to be contained in the source code. Therefore, the necessity for a semantically soundly linked and structured code data management system is a major challenge in the field of software engineering. This paper investigates a domain ontology-based automatic knowledge graph creation method for C# source code. The semantic web, open-source developers, knowledge management, expert systems, and online communities are just a few of the fields where software engineers may now understand and analyze code in a semantic manner. By layering conditional random fields on top of a trained Bi-LSTM network, candidate terms for concepts or entities were extracted.The models were automatically trained on a labeled data corpus while also being manually defined. To improve the classification of terms in a particular source code, BI-LSTM and CRF are integrated. Other characteristics to be extracted from the source code were defined in addition to the basic CRF features, which helped the model understand the categorization constraints. Then, the Bi-LSTM model was utilized to extract relations (taxonomic and non-taxonomic). Max pooling has been used to integrate the links between concepts at the word and code levels. Studies demonstrating the applicability and practicality of the proposed approach make use of the SNIPS-NLU library, a C# library for natural language processing. The evaluation process made use of both expert evaluation and the gold standard ontology that was established by experts. According to an expert analysis of the experiment's results, this approach generated an average f-measure and relevance of 77.04 and 81.275, respectively. By extracting elements and relations from C# and other programming languages that are similar, recurrent neural networks appear to be efficient and promising.