FastTagRec: fast tag recommendation for software information sites

Jin Liu, Pingyi Zhou, Zijiang Yang, Xiao Liu, John Grundy

Research output: Contribution to journalArticleResearchpeer-review

8 Citations (Scopus)

Abstract

Software information sites such as StackOverflow and Freeecode enable information sharing and communication for developers around the world. To facilitate correct classification and efficient search, developers need to provide tags for their postings. However, tagging is inherently an uncoordinated process that depends not only on developers’ understanding of their own postings but also on other factors, including developers’ English skills and knowledge about existing postings. As a result, developers keep creating new tags even though existing tags are sufficient. The net effect is an ever increasing number of tags with severe redundancy along with more postings over time. Any algorithms based on tags become less efficient and accurate. In this paper we propose FastTagRec, an automated scalable tag recommendation method using neural network-based classification. By learning existing postings and their tags from existing information, FastTagRec is able to very accurately infer tags for new postings. We have implemented a prototype tool and carried out experiments on ten software information sites. Our results show that FastTagRec is not only more accurate but also three orders of magnitude faster than the comparable state-of-the-art tool TagMulRec. In addition to empirical evaluation, we have also conducted an user study which successfully confirms the usefulness of of our approach.

Original languageEnglish
Pages (from-to)675-701
Number of pages27
JournalAutomated Software Engineering
Volume25
Issue number4
DOIs
Publication statusPublished - Dec 2018

Keywords

  • Software information site
  • Software object
  • Tag recommendation

Cite this