Article Preview
TopIntroduction
The volume of scientific articles has increased so dramatically in the past decade; it is impossible for a researcher or student to digest all the new information available in the scientific repository. Various online academic service providers have allowed users to access papers through their search engines, such as Google Scholar, ACM Digital Library, ScienceDirect, IEEE Xplore, and Semantic Scholar, where users can input queries to search for relevant articles in their database. Paper recommender systems can complement the search engine, and the possible recommendation scenarios can be grouped into three categories based on the recommendation timing, i.e., before, during, or after a search session (Li et al., 2019). In particular, citation recommendation is provided during a search session, in which related paper recommendations can be displayed beside the content that the user is currently browsing. While citation recommendation falls into one of the scenarios of paper recommendation, there are fundamental differences between citation and paper recommendation. Paper recommendation focuses on providing users with articles that are worthwhile to read and examine in the context of a research topic. Furthermore, a paper recommender system gives personalized article recommendations based on modeling the user’s profile from their behavior history, e.g., clicked/bookmarked/written documents. However, citation recommendation refers to the task of recommending appropriate citations for a text passage within a document (Färber & Jatowt, 2020). While there are some published surveys on paper recommendation (Beel et al., 2016; Bai et al., 2019), the present survey focuses on citation recommendation.
The traditional process of finding relevant citations requires much tedious and manual work. A researcher needs to record a document collection on their own, and the chance of identifying papers for citing depends on whether the researcher already knows the candidates. Another option for acquiring citation candidates is to rely on a bibliographic database, such as Google Scholar, or domain-specific platforms such as DBLP in computer science or PubMed in biomedical and the life sciences domain. However, finding the appropriate query keywords to search for cited papers is skillful work and requires considerable time and effort (Färber & Jatowt, 2020). The motivation for citation recommendation is to provide an efficient method for the citation process. The user provides the written text to the recommender system, and then the system presents a snippet with citations. For instance, given the citation context “FM has a uniform weight in feature interactions; the researchers introduce attention mechanism to enable features that contribute differently for link prediction” within a document, the citation recommendation system might produce two citations as follows: “FM (Rendle, 2010) has a uniform weight in feature interactions; the researchers introduce attention mechanism (Vaswani et al., 2017) to enable features that contribute differently for link prediction.”, where the two corresponding references are added respectively to (1) a publication introducing Factorization Machine (FM) which is a popular technique in modeling feature interaction, and (2) a publication backing up the statement that the proposed method introduces an augmented component for the task of link prediction.
Two user groups can benefit from a citation recommender system, as shown in the following examples:
- •
Expert Group: An experienced researcher familiar with their research field is going to conduct a survey on a research project related to AI education systems. Citation recommendation can be beneficial, as such a user might not be familiar with publications about educational psychology.
- •
Non-expert Group: Newcomers, such as early-stage master’s students and Ph.D. students, might get frustrated when facing the tremendous amount of research publications, and they are unaware of most of the relevant literature in their research areas. Citation recommender systems help those novice researchers to find cite-worthy publications to write their research proposals.