Towards a Reinforcement Learning-based Exploratory Search for Mashup Tag Recommendation

The rapid increase in the number of online mashups requires better management and organization to facilitate mashup discovery, selection and recommendation. Tagging is one of the widely known and efficient ways to better manage and organize web services. Existing tagging methods are typically manual and tedious processes. On the other hand, current automatic tagging methods are either not very effective in describing their associated mashups or omit vital keywords that would enrich the set of candidates for tag selection. This paper presents a reinforcement learning (RL) method to automatically recommend tags for mashups. Our proposed method carries out effective exploratory actions to automatically extract suitable combinations of tags for mashups, through word vector similarities. Using our proposed method in a RL setup, we carry out experiments in an online mashup platform and evaluate our method with a real-world dataset from ProgrammableWeb¹. Our method shows improved performance compared with state-of-the-art baselines.