Context aware code snippet recommendation
View Patent ↗A context aware code snippet recommendation method, system, and computer program product that includes generating comments for source code, learning semantically meaningful sentence embeddings based on the comments, and retrieving code snippets from the source code based on the learned semantically meaningful sentence embeddings.
1. A computer-implemented context aware code snippet recommendation method, the method comprising:
ingesting a snippet of source code characters by an encoder-decoder model, wherein the encoder-decoder converts the characters to vectors;
translating, by the encoder-decoder model, the vectors into natural language comments;
learning semantically meaningful sentence embeddings based on the natural language comments and a comparison between the learned semantically meaningful sentence embeddings and a developer-provided comment that is converted into a sentence embedding, wherein the semantically meaningful sentence embeddings are learned by:
building a code file as a map between each code snippet of the source code and a corresponding generated natural language comment;
converting the code file to a natural language document; and
calculating the semantically meaningful sentence embeddings as vectors by inputting the code file through a Sentence-Bidirectional Encoder Representations from Transformers (BERT) model; and
retrieving code snippets from the source code based on the learned semantically meaningful sentence embeddings, wherein the code snippets are retrieved based on a comparison between the learned semantically meaningful sentence embeddings and a developer-provided comment that is converted into a sentence embedding.
2. The computer-implemented context aware code snippet recommendation method of claim 1 , wherein the comparison between the semantically meaningful sentence embeddings and the sentence embedding of the developer-provided comment is based on the code snippet corresponding to the semantically meaningful sentence embeddings having a largest cosine-similarity to the sentence embedding of the developer-provided comment.
3. The computer-implemented context aware code snippet recommendation method of claim 1 , wherein the comparison between the semantically meaningful sentence embeddings and the sentence embedding of the developer provided comment further includes utilizing project text descriptions in metadata and comments in a current code file that correspond to the developer-provided comment.
4. The computer-implemented context aware code snippet recommendation method of claim 1 , wherein an encoder-decoder model is utilized to generate the comments for the source code with an output of natural language documents including the generated comments.
5. The computer-implemented context aware code snippet recommendation method of claim 4 , wherein the generating repeatedly generates comments for a plurality of source codes to create a repository of the natural language documents.
6. The computer-implemented context aware code snippet recommendation method of claim 5 , wherein the semantically meaningful sentence embeddings include a numerical vector representation of the natural language documents.
7. The computer-implemented context aware code snippet recommendation method of claim 6 , wherein a developer comment for a code snippet is run through a model to generate an embedding including a numerical vector representation of the developer comment, and
wherein the numerical vector representation of the developer comment is compared with the semantically meaningful sentence embeddings in the repository to output a code snippet having the closest cosine-similarity to the numerical vector representation of the developer comment.
8. The computer-implemented context aware code snippet recommendation method of claim 1 , wherein a developer comment for a code snippet is run through the Sentence-Bidirectional Encoder Representations from Transformers (BERT) model to generate an embedding including a numerical vector representation of the developer comment, and
wherein the numerical vector representation of the developer comment is compared with the semantically meaningful sentence embeddings in the repository to output a code snippet having the closest cosine-similarity to the numerical vector representation of the developer comment.
9. The computer-implemented context aware code snippet recommendation method of claim 1 , embodied in a cloud-computing environment.
10. A context aware code snippet recommendation computer program product, the context aware code snippet recommendation computer program product comprising a computer-readable storage medium having program instructions embodied therewith, the program instructions executable by a computer to cause the computer to perform:
ingesting a snippet of source code characters by an encoder-decoder model, wherein the encoder-decoder converts the characters to vectors;
translating, by the encoder-decoder model, the vectors into natural language comments;
learning semantically meaningful sentence embeddings based on the comments, wherein the semantically meaningful sentence embeddings are learned by:
building a code file as a map between each code snippet of the source code and a corresponding generated natural language comment;
converting the code file to a natural language document; and
calculating the semantically meaningful sentence embeddings as vectors by inputting the code file through a sentence-Bidirectional Encoder Representations from Transformers (BERT) model; and
retrieving code snippets from the source code based on the learned semantically meaningful sentence embeddings, wherein the code snippets are retrieved based on a comparison between the learned semantically meaningful sentence embeddings and a developer-provided comment that is converted into a sentence embedding.
11. The context aware code snippet recommendation computer program product of claim 10 , wherein the comparison between the semantically meaningful sentence embeddings and the sentence embedding of the developer-provided comment is based on the code snippet corresponding to the semantically meaningful sentence embeddings having a largest cosine-similarity to the sentence embedding of the developer-provided comment.
12. The context aware code snippet recommendation computer program product of claim 10 , wherein the comparison between the semantically meaningful sentence embeddings and the sentence embedding of the developer-provided comment further includes utilizing project text descriptions in metadata and comments in a current code file that correspond to the developer-provided comment.