Context-based word embedding for programming artifacts

Patent metadata

Field	Value
Publication number	US-11422798-B2
Application number	US-202016801218-A
Country	US
Kind code	B2
Filing date	Feb 26, 2020
Priority date	Feb 26, 2020
Publication date	Aug 23, 2022
Grant date	Aug 23, 2022

Abstract

Techniques for context-based word embedding for programming artifacts are described herein. An aspect includes determining a plurality of keywords based on a corpus of programming artifacts, the corpus of programming artifacts including source code corresponding to a software project. Another aspect includes determining a plurality of context/keyword pair sets based on the plurality of keywords and the corpus of programming artifacts, wherein each context/keyword pair set of the plurality of context/keyword pair sets includes a first keyword, a second keyword, and a context type corresponding to a co-occurrence of the first keyword and the second keyword in the corpus of programming artifacts. Another aspect includes constructing a word embedding matrix based on the plurality of context/keyword pair sets.

First claim

What is claimed is:

1. A computer-implemented method comprising:
determining, by a processor, a plurality of keywords based on a corpus of programming artifacts, the corpus of programming artifacts comprising source code corresponding to a software project;
determining a plurality of context/keyword pair sets based on the plurality of keywords and the corpus of programming artifacts, wherein each context/keyword pair set of the plurality of context/keyword pair sets comprises a first keyword, a second keyword, and a context type corresponding to a co-occurrence of the first keyword and the second keyword in the corpus of programming artifacts; and
constructing a word embedding matrix based on the plurality of context/keyword pair sets,
wherein constructing the word embedding matrix based on the plurality of context/keyword pair sets comprises:
training a latent embedding matrix based on the plurality of context/keyword pair sets;
creating a manifest feature vector based on the plurality of context/keyword pair sets and the corpus of programming artifacts; and
combining the latent embedding matrix with the manifest feature vector corresponding to the corpus of programming artifacts to construct the word embedding matrix, wherein the combining includes ranking the plurality of context/keyword pair sets based on the manifest feature vector; and
wherein the word embedding matrix is used to train a recurrent neural network (RNN) to process source code.

2. The method of claim 1 , wherein determining the plurality of keywords comprises:
determining a naming convention of the corpus of programming artifacts;
determining a plurality of tokens based on the determined naming convention;
constructing a manifest feature vector based on the plurality of tokens and the corpus of programming artifacts;
ranking the plurality of tokens based on the manifest feature vector; and
selecting a subset of the plurality of tokens as keywords based on the manifest feature vector.

3. The method of claim 2 , wherein the naming convention comprises one of camel case, kebab case, and snake case.

4. The method of claim 1 , wherein the context type corresponds to a type of a statement in the source code of the corpus of programming artifacts, wherein the first keyword and the second keyword co-occur in the statement.

5. The method of claim 1 , wherein the context type corresponds to a business rule corresponding to the corpus of programming artifacts, wherein the first keyword and the second keyword co-occur in the business rule.

6. The method of claim 1 , wherein the context type corresponds to a common prefix or suffix of the first keyword and the second keyword in the corpus of programming artifacts.

7. A system comprising:
a memory having computer readable instructions; and
one or more processors for executing the computer readable instructions, the computer readable instructions controlling the one or more processors to perform operations comprising:
determining a plurality of keywords based on a corpus of programming artifacts, the corpus of programming artifacts comprising source code corresponding to a software project;
determining a plurality of context/keyword pair sets based on the plurality of keywords and the corpus of programming artifacts, wherein each context/keyword pair set of the plurality of context/keyword pair sets comprises a first keyword, a second keyword, and a context type corresponding to a co-occurrence of the first keyword and the second keyword in the corpus of programming artifacts; and

constructing a word embedding matrix based on the plurality of context/keyword pair sets,
wherein constructing the word embedding matrix based on the plurality of context/keyword pair sets comprises:
training a latent embedding matrix based on the plurality of context/keyword pair sets; and
creating a manifest feature vector based on the plurality of context/keyword pair sets and the corpus of programming artifacts; and
combining the latent embedding matrix with the manifest feature vector corresponding to the corpus of programming artifacts to construct the word embedding matrix, wherein the combining includes ranking the plurality of context/keyword pair sets based on the manifest feature vector; and
wherein the word embedding matrix is used to train a recurrent neural network (RNN) to process source code.

8. The system of claim 7 , wherein determining the plurality of keywords comprises:
determining a naming convention of the corpus of programming artifacts;
determining a plurality of tokens based on the determined naming convention;
constructing a manifest feature vector based on the plurality of tokens and the corpus of programming artifacts;
ranking the plurality of tokens based on the manifest feature vector; and
selecting a subset of the plurality of tokens as keywords based on the manifest feature vector.

9. The system of claim 8 , wherein the naming convention comprises one of camel case, kebab case, and snake case.

10. The system of claim 7 , wherein the context type corresponds to a type of a statement in the source code of the corpus of programming artifacts, wherein the first keyword and the second keyword co-occur in the statement.

11. The system of claim 7 , wherein the context type corresponds to a business rule corresponding to the corpus of programming artifacts, wherein the first keyword and the second keyword co-occur in the business rule.

12. The system of claim 7 , wherein the context type corresponds to a common prefix or suffix of the first keyword and the second keyword in the corpus of programming artifacts.

13. A computer program product comprising a computer readable storage medium having program instructions embodied therewith, the program instructions executable by one or more processors to cause the one or more processors to perform operations comprising:
determining a plurality of keywords based on a corpus of programming artifacts, the corpus of programming artifacts comprising source code corresponding to a software project;
determining a plurality of context/keyword pair sets based on the plurality of keywords and the corpus of programming artifacts, wherein each context/keyword pair set of the plurality of context/keyword pair sets comprises a first keyword, a second keyword, and a context type corresponding to a co-occurrence of the first keyword and the second keyword in the corpus of programming artifacts; and
constructing a word embedding matrix based on the plurality of context/keyword pair sets,
wherein constructing the word embedding matrix based on the plurality of context/keyword pair sets comprises:
training a latent embedding matrix based on the plurality of context/keyword pair sets; and
creating a manifest feature vector based on the plurality of context/keyword pair sets and the corpus of programming artifacts; and
combining the latent embedding matrix with the manifest feature vector corresponding to the corpus of programming artifacts to construct the word embedding matrix, wherein the combining includes ranking the plurality of context/keyword pair sets based on the manifest feature vector; and
wherein the word embedding matrix is used to train a recurrent neural network (RNN) to process source code.

14. The computer program product of claim 13 , wherein determining the plurality of keywords comprises:
determining a naming convention of the corpus of programming artifacts;
determining a plurality of tokens based on the determined naming convention;
constructing a manifest feature vector based on the plurality of tokens and the corpus of programming a

Assignees

• IBM

Inventors

• Luo Yan
• Shao Liujia
• Xu Yan
• Fan Sibin

Classifications

• G06N3/045

  Combinations of networks · CPC title

• G06N3/044

  Recurrent networks, e.g. Hopfield networks · CPC title

• G06N3/0442

  characterised by memory or gating, e.g. long short-term memory [LSTM] or gated recurrent units [GRU] · CPC title

• G06N3/09

  Supervised learning · CPC title

• G06F8/73 Primary

  Program documentation · CPC title