Bi-Directional LSTM-Based Search Engine for Source Code Retrieval

Abstract

The code retrieval process looks for the most relevant code fragments. For this purpose, many code-search methods support natural-language queries. But still, there are problems with figuring out the pros and cons of each technique and choosing the best one for the task of code search. The proposed approach employs Word2Vec embedding and Bi- LSTM, a neural network model, to identify query title similarities with the existing dataset to search the code fragment. A novel model named BiD-CSE (Bi-Direction LSTM Code Search Engine) is proposed which generates source code recommendations. The proposed system assists programmers in locating appropriate implementations of sample code based on requirements specified in the query. The evaluation of the proposed framework is performed using three datasets. BiD-CSE collects and analyzes the input query which is then parsed to find the best match in the dataset. BiD-CSE finds the desired code by matching variables, functions, doc-strings, and comments. The BiD-CSE model is implemented using a web-based platform that allows users to enter a query and obtain the top ten most effective results. Evaluation results show that the performance of the model is better than the current baseline approaches.