KG2Lib: knowledge-graph-based convolutional network for third-party library recommendation

In the process of software system evolution, software users constantly put forward a large number of expectations. For these expectations, software developers usually use the existing third-party libraries and other software resources to accelerate their development processes. At present, tons of th...

Full description

Saved in:
Bibliographic Details
Published inThe Journal of supercomputing Vol. 79; no. 1; pp. 1 - 26
Main Authors Zhao, Jing-zhuan, Zhang, Xuan, Gao, Chen, Li, Zhu-dong, Wang, Bao-lei
Format Journal Article
LanguageEnglish
Published New York Springer US 2023
Springer Nature B.V
Subjects
Compilers
Computer Science
Interpreters
Knowledge representation
Libraries
Performance evaluation
Processor Architectures
Programming Languages
Recommender systems
Software development
Third party
Library recommendation
Graph convolutional network
Software development
Knowledge graph
Online AccessGet full text

Cover

More Information
Summary:In the process of software system evolution, software users constantly put forward a large number of expectations. For these expectations, software developers usually use the existing third-party libraries and other software resources to accelerate their development processes. At present, tons of third-party libraries are available. Therefore, appropriate recommendation methods are very important for developers to find suitable libraries for their development projects. In this paper, we present KG2Lib, a recommendation method to assist software developers in selecting suitable software libraries for their current projects. KG2Lib exploits a knowledge-graph-based convolutional network to recommend software libraries by relying on a set of libraries which were already called by current projects. The interaction matrix, weight matrix and knowledge graph are the inputs of KG2Lib. What’s more, KG2Lib recommends libraries to developers from project level and library level, which can better capture the fine-grained information to achieve better recommend performance. The performance of KG2Lib was evaluated on three datasets with four existing baseline models. The experimental results show that KG2Lib achieves better performance and helps software developers accurately select the appropriate third-party libraries.
ISSN:0920-8542
1573-0484
DOI:10.1007/s11227-022-04603-3