Abstract:
In software development, the availability of useful and adaptable programming components
or source codes is crucial. Traditional information retrieval techniques fall short in
code search, as these require bridging the semantic gap between source code and natural
language based queries for search. This dissertation tackles the challenge of code search
in software development by offering a code retrieval framework that offers solutions based
on ontologies, machine learning, and deep learning techniques. The proposed framework
uses ontologies for source code search, a machine learning-based ranking schema, and
advanced methods such as graph neural networks and Bi-LSTM-based neural attention.
The evaluation results demonstrates the effectiveness of our approach through extensive
experimentation with benchmark datasets to produce improved performance compared
to existing methods. Based on our results, we can claim that software developers who
want to speed up development and reduce the development cost can use the proposed
framework.
