Context-based recommendation

Abstract

With the rapid growth of the scientific literature, citation recommendation systems able to speed up literature review and citing process during a research process. Recent approaches use bag-of-word retrieval to represent the documents, which discards word order information which is important in representation for documents. This project presents a method of recommend candidate references using document representations based on context of each document by learning document representations that incorporate inter-document document relatedness using citation graph and the state-of-the-art Transformer language model. Documents can be embedded into a high-dimensional vector space. Given a query document, it can be encoded into a vector which its nearest neighbours could be retrieved as candidates for citation. A recommendation web application is implemented to facilitate the citation recommendation.