QA4GIS: A novel approach learning to answer GIS developer questions with API documentation
Document Type
Article
Publication Date
10-1-2021
Abstract
Community-based question answering websites have attracted more and more scholars and developers to discuss domain knowledge and software development. In this article, we focus on the GIS section of the Stack Exchange website and develop a novel approach, QA4GIS, a deep learning-based system for question answering tasks with a deep neural network (DNN) model to extract the representation of the query–API document pair. We use the LambdaMART model to rerank the candidate API documents. We begin with an empirical analysis of the questions and answers, demonstrating that API documents could answer 52.93% of the questions. Then we evaluate QA4GIS by comparing it with 10 other baselines. The experiment results show that QA4GIS can improve 21.39% on the MAP score and 22.34% on the MRR score compared with the best baseline SIF.
Identifier
85110425806 (Scopus)
Publication Title
Transactions in GIS
External Full Text Location
https://doi.org/10.1111/tgis.12798
e-ISSN
14679671
ISSN
13611682
First Page
2675
Last Page
2700
Issue
5
Volume
25
Grant
1937908
Recommended Citation
Wang, Wenbo; Li, Yi; Wang, Shaohua; and Ye, Xinyue, "QA4GIS: A novel approach learning to answer GIS developer questions with API documentation" (2021). Faculty Publications. 3767.
https://digitalcommons.njit.edu/fac_pubs/3767