A knowledge graph approach to mashup tag recommendation

Tags have been extensively used to organize and index mashup services. However, the selection of relevant tags that depict functionality of mashups has remained a daunting task. This is because mashups have different functionalities than their constituent web APIs. Some existing tag recommendation methods usually follow a manual approach, which is time consuming and prone to errors. Others propose some means of automatic tag recommendation that use a similarity measure which has to be re-computed for every new mashup against the entire mashup and web API database. Such methods are also time consuming, inefficient and therefore not practical. In this paper, we present an automatic tag recommendation method for mashups, using knowledge graphs (KG). The method uses as entry points (seeds) into the KG, topics from mashup description, its primary category, and its constituent web APIs. From the seeds, we walk the graph to extract candidate tags based on node cosine similarity. We finally employ word similarity as a scoring function to explore and rank the candidate tags. Top-ranked candidate tags are subsequently recommended. We conduct experiments, with a real world dataset from programmable web1, and compare our results to existing baselines. Our results show that our model outperforms the baselines in all cases.