Portal de Periódicos da CAPES

TagDC: A tag recommendation method for software information sites with a combination of deep learning and collaborative filtering

2020; Elsevier BV; Volume: 170; Linguagem: Inglês

10.1016/j.jss.2020.110783

1873-1228

Can Li, Ling Xu, Meng Yan, Yan Lei,

Topic Modeling

Software information sites (e.g., StackOverflow, Freecode, etc.) are increasingly essential for software developers to share knowledge, communicate new techniques, and collaborate. With the rapid growth of software objects, tags are widely applied to aid developers' various operations on software information sites. Since tags are freely and optionally selected by developers, the differences in background, expression habits, and understanding of software objects among developers may cause inconsistent or inappropriate tags. To alleviate the problems of tag synonyms and tag explosion, we propose TagDC, i.e., a composite Tag recommendation method with Deep learning and Collaborative filtering. TagDC consists of two complementary modules: the word learning enhanced CNN capsule module (TagDC-DL) and the collaborative filtering module (TagDC-CF). It can improve the understanding of software objects from different perspectives. Given a new software object, TagDC can calculate a list of the combined confidence probabilities of tags and then recommend TOP-K tags by ranking the probabilities in the list. We evaluated our TagDC on nine datasets with different scales. The experimental results show that TagDC achieves a better effectiveness against two state-of-the-art baseline methods (i.e., TagCNN and FastTagRec) with a substantial improvement.