Learning Semantic Vector Representations of Source Code via a Siamese Neural Network

Bo Zhang; David Wehr; Eleanor Pence; Guilherme Ferreira; Halley Fede; John Walczyk; Joseph Hughes

RECORD

Learning Semantic Vector Representations of Source Code via a Siamese Neural Network

David Wehr, Halley Fede, Eleanor Pence, Bo Zhang, Guilherme Ferreira, John Walczyk, Joseph Hughes

Machine Learning · Submitted 26 Apr 2019

abstract compiled · 2026-09-23 · no language-model review

The authors' main claim Our model significantly outperforms a bag-of-tokens embedding, providing promising results for improving code embeddings that can be used in future software engineering tasks.

how it is stated · asserted 0.96 read the record →

Comment (0) Read on arXiv ↗

Cite ▾

arXiv ID1904.11968 (version 1)

Pith ID2NEKVEI2, Pith's permanent record of this paper and its review

CitationBo Zhang, David Wehr, Eleanor Pence, Guilherme Ferreira, Halley Fede, John Walczyk, Joseph Hughes. "Learning Semantic Vector Representations of Source Code via a Siamese Neural Network." arXiv:1904.11968 (2019). https://pith.science/paper/2NEKVEI2

Claim paper

The compiled record

The paper's own sentences, arranged. Each label is drawn from a closed vocabulary; the number beside it is the probability the classifier gave that label. Numbers in the tables are the authors', checked by code. No sentence in this record was written by a model.

Open the compiled record

Compiled by classifier-1.13.0 via TypeSafe API, question bank abstract-p1-20260923, 2026-09-23: 1 request, 3,581 input tokens.

What the abstract itself states, as probabilities: claims first 0.04 · states limitation 0.07 · human participants 0.03 · controlled comparison 0.90 · randomized assignment 0.01 · states quantitative result 0.07 · compares with prior results 0.95 · mentions code or data availability 0.03.

The abstract, sentence by sentence

arXiv listing abstract v1. Results and the main claim are underlined; show every role.

The abundance of open-source code, coupled with the success of recent advances in deep learning for natural language processing, has given rise to a promising new application of machine learning to source code. In this work, we explore the use of a Siamese recurrent neural network model on Python source code to create vectors which capture the semantics of code. We evaluate the quality of embeddings by identifying which problem from a programming competition the code solves. Our model significantly outperforms a bag-of-tokens embedding, providing promising results for improving code embeddings that can be used in future software engineering tasks.

background contribution method result interpretation main claim

What the classifier reports for each sentence. The number is the probability it gave the label.
#	sentence	role	the authors' own work	states the main conclusion	strength
s0	The abundance of open-source code, coupled with the success of recent advances in deep learning for natural language processing, has given rise to a promising new application of machine learning to source code.	background 1.00	0.04	0.03	asserted 0.85
s1	In this work, we explore the use of a Siamese recurrent neural network model on Python source code to create vectors which capture the semantics of code.	method 0.54	0.94	0.05	asserted 0.83
s2	We evaluate the quality of embeddings by identifying which problem from a programming competition the code solves.	method 0.94	0.94	0.04	asserted 0.95
s3	Our model significantly outperforms a bag-of-tokens embedding, providing promising results for improving code embeddings that can be used in future software engineering tasks.	result 0.82	0.95	0.88	asserted 0.96

What kind of paper the classifier reports this is: method or tool 0.97 · evidence from observation or experiment 0.94.

What was examined, and what was not

sources read

arXiv listing abstract v1

not examined

full text
figures and tables
references and citing sentences
relations to other papers

levels run

P abstract

not run

S sections
T body sentences
C citations
R cross-paper relations

Every quoted sentence on this record is the authors'. Every label is from Pith's closed vocabulary, with the probability shown beside it. Every number in the measurements table is the authors', checked by code. Record as JSON

The paper

Original abstract

The abundance of open-source code, coupled with the success of recent advances in deep learning for natural language processing, has given rise to a promising new application of machine learning to source code. In this work, we explore the use of a Siamese recurrent neural network model on Python source code to create vectors which capture the semantics of code. We evaluate the quality of embeddings by identifying which problem from a programming competition the code solves. Our model significantly outperforms a bag-of-tokens embedding, providing promising results for improving code embeddings that can be used in future software engineering tasks.

Human review

How Pith checks papers → · Full audit of this paper →

How the literature uses it

What this paper does with the arXiv papers it cites

Each citing sentence is taken from the citing paper's LaTeX source and typed by Pith's classifier. A label appears at probability 0.8 or more (0.95 for support), a contest at 0.7 or more; the quoted words are the citing authors'. How each label was checked · JSON

What it cites

Uses the method or tool of 3 papers

For pre-training, we collected all Python projects on public GitHub [ref] that had over 100 stars (using code from [ref] ) and extracted their functions as separate training samples, resulting in 1.3 million total samples.

Learning Python Code Suggestion with a Sparse Pointer Network · Data · 0.94
Because labeled training data is hard to obtain (requiring independent implementations), we pre-train the network by adding a decoder RNN after the encoding stage, turning it into a sequence-to-sequence model [ref] .

Sequence to Sequence Learning with Neural Networks · Pre-training · 0.95
Both the pre-training and final model were trained using the Adam [ref] weight-update algorithm using standard backpropagation techniques.

Adam: A Method for Stochastic Optimization · Pre-training · 1.00

Cites as background 5 papers

Alon et al. [ref] learn code vector embeddings by evaluating paths in the AST and evaluate the resulting vectors by predicting method names from code snippets.

code2vec: Learning Distributed Representations of Code · Related Work · 0.83
The work of Mout et al. [ref] encodes nodes of an AST using a weighted mixture of left and right weight matrices and children of that node.

Convolutional Neural Networks over Tree Structures for Programming Language Processing · Related Work · 0.84
More complex loss functions for Siamese neural networks have been developed, many of which enforce only a margin between classes [ref] , and some which use a triplet of (anchor, positive sample, negative sample) [ref] .

FaceNet: A Unified Embedding for Face Recognition and Clustering · Proposed Model · 0.98

Discussion

Know something about this paper? Leave a comment, or highlight any passage above to comment on it.