Dual-interactive fusion for code-mixed deep representation learning in tag recommendation

Abstract

Automatic tagging on software information sites is a tag recommendation service. It aims to recommend content-based tags for a software object to help developers make distinctions among software objects. Due to deep correlations between software objects and tags, it is challenging to simultaneously consider the code snippet and text description of a software object. Towards automatic tagging service, we propose a novel CDR4Tag method, Code-mixed Deep Representation learning via dual-interactive fusion forTag recommendation on software information sites. In CDR4Tag, a code-mixed dual interaction strategy is designed to fuse the deep semantic correlations between software objects and tags into a joint representation space. On the basis of it, the matching probability is predicted to complete our tag recommendation. Comprehensive experimental results on four software information site datasets have demonstrated the effectiveness of our proposed CDR4Tag in tag recommendation compared with the state-of-the-art methods.