Granted Patent B2

US 12,475,318 · App. 17/814,518 · Granted Nov 18, 2025

Inventors: Mehdi Bahrami (San Jose, CA); Wei-Peng Chen (Fremont, CA)

Assignee: Fujitsu Limited

G06F40/30G06F8/30G06F8/36G06F8/42G06F8/436G06F8/65G06F8/73G06F11/3624G06F16/951G06F18/22G06F18/23213G06F18/285G06F40/166G06F40/211G06F40/216G06F40/242G06F40/40G06F40/44G06N3/04G06N3/08

Abstract

Claims (77)

1 . A method, executed by a processor, comprising:

receiving a source code file associated with a software package;

generating an abstract syntax tree (AST) of at least a portion of a computer-executable code included in the source code file, the computer executable code including metadata describing functions of the computer executable code;

determining a set of functions from the computer-executable code, based on the AST;

extracting the metadata associated with the set of functions from the computer-executable code;

determining one or more parameters associated with each of the set of functions;

determining one or more return values associated with each of the set of functions;

detecting a presence of the one or more parameters and the one or more return values in a description of a corresponding function in the metadata;

selecting, from the set of functions based on the detecting, a subset of functions for which descriptions in the extracted metadata satisfy filtering criteria, the filtering criteria requires a description of the selected subset of functions be included in the extracted metadata;

updating the computer-executable code by filtering lines of code (LoCs) corresponding to the subset of functions from the computer-executable code;

generating a dataset of code features and respective metadata features based on the updated computer-executable code and the metadata; and

training a language model on a sequence-to-sequence generation task, based on the dataset.

2 . The method according to claim 1 , further comprising:

receiving a repository address of a package dataset as an input;

extracting a list of software packages from a data source by using the repository address;

selecting the software package from the extracted list of software packages; and

retrieving, from the data source, resource files associated with the selected software package, wherein the retrieved resource files include the source code file.

3 . The method according to claim 1 , further comprising identifying code components from the computer-executable code by using the AST,

wherein the code components include one or more of: classes, methods included in the classes, and functions different from the methods, and

wherein the set of functions corresponds to the methods and the functions included in the code components.

4 . The method according to claim 1 , wherein the metadata comprises at least one of a function name, a count of parameters associated with a function, a count of one or more default parameters associated with the function, a default value of the one or more default parameters, a type of each of the parameters, a description of the function, or one or more decorators associated with the function.

5 . The method according to claim 4 , wherein the description comprises at least one of a description of each of the parameters, a description of a return value associated with the function, a description of the one or more default parameters associated with the function, or a description of an objective of the function.

6 . The method according to claim 1 , wherein the filtering criteria specifies rules of selecting a function from the set of functions, and

the rules include a first rule to determine whether a description is present in the metadata and a second rule to determine whether the description in the metadata includes:

a description of one or more parameters associated with the function, and

a description of one or more return values associated with the function.

7 . The method according to claim 1 , further comprising:

determining nodes corresponding to the subset of functions in the AST;

selecting a node from the determined nodes;

marking the selected node for a pruning operation based on a determination that the selected node corresponds to a code component that is different from a function call with one or more parameters; and

obtaining a revised AST by executing the pruning operation to remove the marked node from the AST.

8 . The method according to claim 7 , wherein the computer-executable code is updated further by:

extracting, from the computer-executable code, a sequence of function calls corresponding to each function of the subset of functions by using the revised AST; and

generating abstracted code data by abstracting the sequence of function calls.

9 . The method according to claim 8 , further comprising determining return values corresponding to each function of the subset of functions from the updated computer-executable code.

10 . The method according to claim 9 , wherein the generated dataset is a dictionary in a key-value format and the dataset includes:

the code features that include the return values and the abstracted code data, and

the metadata features which include respective natural language (NL) text portions of the descriptions corresponding to the subset of functions, class names corresponding to the subset of functions, function names corresponding to the subset of functions, and a path of the source code file.

11 . The method according to claim 1 , wherein the sequence-to-sequence generation task is a code synthesis task, a code retrieval task, or a software package analysis task.

12 . The method according to claim 1 , further comprising:

receiving an input associated with a natural language query; and

generating lines of computer-executable code based on application of the trained language model on the received input.

13 . A non-transitory computer-readable storage medium configured to store instructions that, in response to being executed, causes a system to perform operations, the operations comprising:

receiving a source code file associated with a software package;

generating an abstract syntax tree (AST) of at least a portion of a computer-executable code included in the source code file, the computer executable code including metadata describing functions of the computer executable code;

determining a set of functions from the computer-executable code, based on the AST;

extracting metadata associated with the set of functions from the computer-executable code;

determining one or more parameters associated with each of the set of functions;

determining one or more return values associated with each of the set of functions;

detecting a presence of the one or more parameters and the one or more return values in a description of a corresponding function in the metadata;

selecting, from the set of functions based on the detecting, a subset of functions for which descriptions in the extracted metadata satisfy filtering criteria, the filtering criteria requires a description of the selected subset of functions be included in the extracted metadata;

updating the computer-executable code by filtering lines of code (LoCs) corresponding to the subset of functions from the computer-executable code;

generating a dataset of code features and respective metadata features based on the updated computer-executable code and the metadata; and

training a language model on a sequence-to-sequence generation task, based on the dataset.

14 . The non-transitory computer-readable storage medium according to claim 13 , wherein the operations further comprise:

receiving a repository address of a package dataset as an input;

extracting a list of software packages from a data source by using the repository address;

selecting the software package from the extracted list of software packages; and

retrieving, from the data source, resource files associated with the selected software package, wherein the retrieved resource files include the source code file.

15 . The non-transitory computer-readable storage medium according to claim 13 , wherein the operations further comprise identifying code components from the computer-executable code by using the AST,

wherein the code components include one or more of classes, methods included in the classes, and functions different from the methods, and

wherein the set of functions corresponds to the methods and the functions included in the code components.

16 . The non-transitory computer-readable storage medium according to claim 13 , wherein the metadata comprises at least one of a function name, a count of parameters associated with a function, a count of one or more default parameters associated with the function, a default value of the one or more default parameters, a type of each of the parameters, a description of the function, or one or more decorators associated with the function.

17 . The non-transitory computer-readable storage medium according to claim 16 , wherein the description comprises at least one of a description of each of the parameters, a description of a return value associated with the function, a description of the one or more default parameters associated with the function, or a description of an objective of the function.

18 . A system, comprising:

a processor configured to:

receive a source code file associated with a software package;

generate an abstract syntax tree (AST) of at least a portion of a computer-executable code included in the source code file, the computer executable code including metadata describing functions of the computer executable code;

determine a set of functions from the computer-executable code, based on the AST;

extract metadata associated with the set of functions from the computer-executable code;

determine one or more parameters associated with each of the set of functions;

determine one or more return values associated with each of the set of functions;

detect a presence of the one or more parameters and the one or more return values in a description of a corresponding function in the metadata;

select, from the set of functions based on the detection, a subset of functions for which descriptions in the extracted metadata satisfy filtering criteria, the filtering criteria requires a description of the selected subset of functions be included in the extracted metadata;

update the computer-executable code by filtering lines of code (LoCs) corresponding to the subset of functions from the computer-executable code;

generate a dataset of code features and respective metadata features based on the updated computer-executable code and the metadata; and

train a language model on a sequence-to-sequence generation task, based on the dataset.

Assignments (1)

Continuity (2)

References Cited (28)