Search

It is crucial for universities to recognize early signals of scientific convergence. Scientific convergence describes a dynamic pattern where the distance between different fields of knowledge shrinks over time. This knowledge space is beneficial to radical innovations and new promising research topics. Research in converging areas of knowledge can therefore allow universities to establish a leading position in the science community. The Q-AKTIV project develops a new approach on the basis of machine learning to identify scientific convergence at an early stage. In this work, we briefly present this approach and the first results of empirical validation. We discuss the benefits of an instrument building on our approach for the strategic management of universities and other research institutes.

Due to its numerous bibliometric entries of scholarly articles and connected information Wikidata can serve as an open and rich source for deep scientometrical analyses. However, there are currently certain limitations: While 31.5% of all Wikidata entries represent scientific articles, only 8.9% are entries describing a person and the number of entries researcher is accordingly even lower. Another issue is the frequent absence of established relations between the scholarly article item and the author item although the author is already listed in Wikidata. To fill this gap and to improve the content of Wikidata in general, we established a workflow for matching authors and scholarly publications by integrating data from the ORCID (Open Researcher and Contributor ID) database. By this approach we were able to extend Wikidata by more than 12k author-publication relations and the method can be transferred to other enrichments based on ORCID data. This is extension is beneficial for Wikidata users performing bibliometrical analyses or using such metadata for other purposes.

We assess the suitability of word embeddings for practical information retrieval scenarios. Thus, we assume that users issue ad-hoc short queries where we return the first twenty retrieved documents after applying a boolean matching operation between the query and the documents. We compare the performance of several techniques that leverage word embeddings in the retrieval models to compute the similarity between the query and the documents, namely word centroid similarity, paragraph vectors, Word Mover’s distance, as well as our novel inverse document frequency (IDF) re-weighted word centroid similarity. We evaluate the performance using the ranking metrics mean average precision, mean reciprocal rank, and normalized discounted cumulative gain. Additionally, we inspect the retrieval models’ sensitivity to document length by using either only the title or the full-text of the documents for the retrieval task. We conclude that word centroid similarity is the best competitor to state-of-the-art retrieval models. It can be further improved by re-weighting the word frequencies with IDF before aggregating the respective word vectors of the embedding. The proposed cosine similarity of IDF re-weighted word vectors is competitive to the TF-IDF baseline and even outperforms it in case of the news domain with a relative percentage of 15%.