Staying aware of new approaches emerging within specific areas can be challenging for researchers who have to follow many feeds such as journals articles, authors' papers, and other basic keyword-based matching algorithms. Hence, this paper proposes an information retrieval process for scientific articles aiming to suggest semantically related articles using exclusively a knowledge base. The first step categorizes articles by the disambiguation of their keywords by identifying common categories within the knowledge base. Then, similar articles are identified using the information extracted from the categorization, such as synonyms. The experimental evaluation shows that the proposed approach significantly outperforms the well known cosine similarity measure of vectors angles inherited from word2vec embeddings. Indeed, there is a difference of 30% for P@k (k ∈ [1, 100]) in favor of the proposed approach.

No abstract