A Hybrid Semantic Relatedness Algorithm by Entity Co-Occurrence and Specialized Word Embeddings

As the amount of biomedical literature increases, measuring semantic relatedness between two entities is one of the important tasks from the perspective of finding meaningful biological relationships. In this paper, we propose a hybrid semantic relatedness algorithm for biomedical knowledge discovery. We incorporate a co-occurrence approach to capture hidden relationship and adopt specialized word embeddings by considering both direct and indirect entity relationships. We analyze and evaluate our proposed method with other well-accepted methods such as co-occurrence, Word2Vec, COALS, and random indexing by comparing top entities related with Alzheimer's disease. In addition, we conduct a series of analyses including gene, pathway, and gene-phenotype relationship analysis.