Discovering Search Space Using M-distance Clustering of Semantic Relatedness Based Weighted Network for the Content-based Recommender System

As part of the research process, relevant documents are identified to keep up with the latest advancements in the domain. Document recommendation systems are used by researchers as a means of accomplishing this goal. Textual content, collaborative filtering, and citation information-based approaches are among the proposed approaches for the recommendation systems. Content-based techniques take advantage of the entire text of papers and produce more promising results, but comparing input document text data to every document in the dataset is not practical for the content-based recommender system. This study looks into the possibility of using bibliographic data to reduce the number of comparisons. The proposed  system is based on the assumption that two scientific papers are semantically connected if they are co-cited more frequently than by chance. The likelihood of co-citation, also known as semantic relatedness, can be used to quantify this connection. This work presents a new way to distribute the weight among connected scholarly documents based on a semantic relatedness score. Our proposed solution eliminates a substantial amount of needless text comparisons for the system by gathering scholarly document pairs with high likelihood values and using them  as a search area for the content-based recommender system. By spreading the co-citation relationship out to certain distances, the proposed approach can find relevant documents that are not found by traditional co-citation searches. The results reveal that the system is capable of reducing computations by a significant margin and of detecting false positive situations in content comparison using Doc2vec.

Keywords: Semantic Relatedness, Content-based Recommended System, Graph Clustering, Co-citation network.