SemiTagRec: A Semi-supervised Learning Based Tag Recommendation Approach for Docker Repositories

Abstract

Docker has been the mainstream technology for providing reusable software artifacts by packaging applications, dependencies, and execution environments into images. Developers can easily build and deploy their applications using Docker. Currently, a large number of reusable Docker repositories are in the online open source communities, especially Docker Hub and Docker Store. Effectively reusing these artifacts requires a well understanding of them, and semantic tags provide this way. However, the communities do not support tags well, and little training data is available. This paper addresses the problem and proposes a semi-supervised learning based tag recommendation approach, SemiTagRec, for Docker repositories. SemiTagRec contains four components. (1) Predictor calculates the probabilities of assigning tags to Docker repositories. (2) Extender introduces in new tags as the candidates based on tag correlation analysis. (3) Evaluator measures the candidate tags. (4) Integrator combines the results of predictor and evaluator, and then takes the tags with high scores as the final result. SemiTagRec uses the newly tagged repositories together with the original ones as training data for the next round of training. In this iterative manner, SemiTagRec trains the predictor with the cumulative labeled data set and the extended tag vocabulary to achieve high accuracy of tag recommendation. Finally, we conducted some experiments and evaluated SemiTagRec by comparing it with other related works. Experimental results show that SemiTagRec outperforms the other approaches in terms of Recall@5 and Recall@10.