Using Word Embeddings for Recommending Datasets based on Scientific Publications

In scholarly search systems, computing recommendations of the same type, for example, additional publications when reading a particular publication, is a well-approached problem. However, suggesting items from another type, e.g., research data when reading a publication, is rarely covered in scholarly recommendations. In this position paper, we employ word embeddings to approach the problem of such cross-domain recommendations in scientific search systems, more specifically, recommending research data based on publications. Besides various metadata, publication and research dataset entries comprise textual metadata (e.g. title, abstract), which allows to detect similar entries using word embeddings. We illustrate first results, major problems and possible solutions when using word embeddings for recommending datasets based on publications.