Exploiting Citation Knowledge in Personalised Recommendation of Recent Scientific Publications

Proceedings of the 12th Conference on Language Resources and Evaluation (LREC 2020), pages 2231–2240 Marseille, 11–16 May 2020 c⃝ European Language Resources Association (ELRA), licensed under CC-BY-NC 2231 Exploiting Citation Knowledge in Personalised Recommendation of Recent Scientific Publications Anita Khadka1, Iv´an Cantador2, Miriam Fernandez1 1Knowledge Media Institute, The Open University, UK 2Escuela Polit´ecnica Superior, Universidad Aut´onoma de Madrid, Spain anita.khadka@open.ac.uk, ivan.cantador@uam.es, miriam.fernandez@open.ac.uk Abstract In this paper we address the problem of providing personalised recommendations of recent scientific publications to a particular user, and explore the use of citation knowledge to do so. For this purpose, we have generated a novel dataset that captures authors’ publication history and is enriched with different forms of paper citation knowledge, namely citation graphs, citation positions, citation contexts, and citation types. Through a number of empirical experiments on such dataset, we show that the exploitation of the extracted knowledge, particularly the type of citation, is a promising approach for recommending recently published papers that may not be cited yet. The dataset, which we make publicly available, also represents a valuable resource for further investigation on academic information retrieval and filtering. Keywords: Research publication dataset, citation context, citation types, recommender systems 1. Introduction Keeping up with the most recent scientific literature is a challenge for many researchers giving the continuous and increasing growth of academic publications. A re- cent report1 claims the existence of 33,100 active scholarly peer-reviewed scientific, technical and medical English- language journals in mid-2018 (plus a further 9,400 non- English-language journals), collectively publishing over 3 million papers a year. The report also states that the pro- duction of scientific publications is steadily increasing at a 4% yearly rate. Aiming to address this information overload problem, aca- demic search engines like Google Scholar2 or PubMed,3, as well as Recommender Systems (RS) (Beel et al., 2016), have emerged in the last years. RS in particular have fo- cused on suggesting relevant papers for a given user (Mid- dleton et al., 2001; Torres et al., 2004), but have also ad- dressed other relevant tasks, including recommending rel- evant papers for a particular snapshot of content, such as title and abstract (Bethard and Jurafsky, 2010; Huang et al., 2015), recommending relevant papers for a given publica- tion (Liang et al., 2011a; Khadka and Knoth, 2018), recom- mending relevant papers for a particular collection of publi- cations (Ekstrand et al., 2010; Shimbo et al., 2007), and rec- ommending relevant papers for an undergoing manuscript, i.e., a paper yet to be published (He et al., 2010; Strohman et al., 2007). Most of academic paper recommender systems, however, do not take into consideration the time when papers were published, and do not address the real-world problem of recommending recently published papers (Ha et al., 2014). In fact, traditional collaborative filtering systems are ineffective to address that problem, since they are not able to recommend the latest, most recent papers, which have 1https://www.stm-assoc.org/2018_10_04_ STM_Report_2018.pdf 2https://scholar.google.co.uk/ 3https://www.ncbi.nlm.nih.gov/pubmed/ not been rated or cited yet. On the other hand, while there are approaches that …

Content truncated to 500 words