Measuring the Semantic World – How to Map Meaning to High-Dimensional Entity Clusters in PubMed?

摘要: The exponential increase of scientific publications in the medical field urgently calls for innovative access paths beyond limits a term-based search. As an example, search term “diabetes” leads to result over 600,000 digital library PubMed. In such cases, automatic extraction semantic relations between important entities like active substances, diseases, and genes can help reveal entity-relationships thus allow simplified knowledge embedded libraries. On other hand, semantic-relation tasks distributional embedding models based on neural networks promise considerable progress terms accuracy, performance scalability. Yet, despite recent successes this field, questions arise related their non-deterministic nature: Are meaningful, perhaps even new unknown entity-relationships? paper, we address question by measuring associations pharmaceutical as substances (drugs) diseases high-dimensional space. our investigation, show that while one hand only few contextualized directly correlate with spatial distance, have discovered potential predicting associations, which makes method suitable new, literature-based technique practical e.g., drug repurposing.