Content-Based Citation Recommendation

Proceedings of NAACL-HLT 2018, pages 238–251 New Orleans, Louisiana, June 1 - 6, 2018. c⃝2018 Association for Computational Linguistics Content-Based Citation Recommendation Chandra Bhagavatula Allen Institute for AI chandrab@allenai.org Sergey Feldman Data Cowboys ∗ sergey@data-cowboys.com Russell Power Independent Researcher † russell.power@gmail.com Waleed Ammar Allen Institute for AI waleeda@allenai.org Abstract We present a content-based method for rec- ommending citations in an academic paper draft. We embed a given query document into a vector space, then use its nearest neighbors as candidates, and rerank the candidates us- ing a discriminative model trained to distin- guish between observed and unobserved cita- tions. Unlike previous work, our method does not require metadata such as author names which can be missing, e.g., during the peer review process. Without using metadata, our method outperforms the best reported results on PubMed and DBLP datasets with relative improvements of over 18% in F1@20 and over 22% in MRR. We show empirically that, although adding metadata improves the per- formance on standard metrics, it favors self- citations which are less useful in a citation rec- ommendation setup. We release an online por- tal for citation recommendation based on our method,1 and a new dataset OpenCorpus of 7 million research articles to facilitate future research on this task. 1 Introduction Due to the rapid growth of the scientific litera- ture, conducting a comprehensive literature review has become challenging, despite major advances in digital libraries and information retrieval sys- tems. Citation recommendation can help improve the quality and efficiency of this process by sug- gesting published scientific documents as likely citations for a query document, e.g., a paper draft to be submitted for ACL 2018. Existing citation recommendation systems rely on various informa- tion of the query documents such as author names and publication venue (Ren et al., 2014; Yu et al., ∗ Work done while on contract with AI2 † Work done while at AI2 1 http://labs.semanticscholar.org/ citeomatic/ 2012), or a partial list of citations provided by the author (McNee et al., 2002; Liu et al., 2015; Jia and Saule, 2017) which may not be available, e.g., during the peer review process or in the early stage of a research project. Our method uses a neural model to embed all available documents into a vector space by encod- ing the textual content of each document. We then select the nearest neighbors of a query document as candidates and rerank the candidates using a second model trained to discriminate between ob- served and unobserved citations. Unlike previous work, we can embed new documents in the same vector space used to identify candidate citations based on their text content, obviating the need to re-train the models to include new published pa- pers. Further, unlike prior work (Yang et al., 2015; Ren et al., 2014), our model is computationally ef- ficient and scalable during both training and test time. We assess the feasibility of recommending cita- tions when some metadata for the query document is missing, and find that we are able to …

Content truncated to 500 words