A Hybrid Approach for Paper Recommendation

Abstract

Paper recommendation has become an increasingly important yet challenging task due to the rapidly expanding volume and scope of publications in the broad research community. Due to the lack of user profiles in public digital libraries, most existing methods for paper recommendation are through paper similarity measurements based on citations or contents, and still suffer from various performance issues. In this paper, we construct a graphical form of citation relations to identify relevant papers and design a hybrid recommendation model that combines both citationand content-based approaches to measure paper similarities. Considering that citations at different locations in one article are likely of different significance, we define a concept of citation similarity with varying weights according to the sections of citations. We evaluate the performance of our recommendation method using Spearman correlation on real publication data from public digital libraries such as CiteSeer and Wanfang. Extensive experimental results show that the proposed hybrid method exhibits better performance than state-of-the-art techniques, and achieves 40% higher recommendation accuracy in average in comparison with citation-based approaches.

Scientific paper recommendation systems: a literature review of recent publications

“… When considering the broadest definition of graph-based methods many recent paper recommendation systems tend to belong to the class of hybrid methods. Most of the approaches [ 5 , 46 , 48 , 49 , 57 , 88 , 105 , 117 ] utilise some type of graph structure information as part of the approach which would classify them as graph-based but as they also utilise historic user-interaction data or descriptions of paper features (see, e.g. Li et al [ 57 ] who describe their approach as network-based while using a graph structure, textual components and user profiles) which would render them as either CF or CBF also.…”

“…Key phrase candidates were weighted and the top 20 represent candidate papers. Kang et al [ 46 ] extract key phrases from CiteSeer to describe the diversity of recommended papers. Renuka et al [ 86 ] apply rapid automatic keyword extraction.…”

“…Kang et al [ 46 ] crawl full texts of papers from CiteSeer and construct citation graphs to determine candidate papers. Then they compute a combination of section-based citation and key phrase similarity to rank recommendations.…”

“…3 sets: data from 2018 and 2019 (221,076 p , 503,945 a ), mathematical analysis (98,702 p , 117,183 a ), image processing (49,098 p , 107,290 a ) AMiner_tiny [ 30 ] 188 input p , 10 candidate p for each input AMiner_huge [ 108 ] 2,092,356 p , 1,712,433 a , 8,024,869 c , 4,258,615 co-autorships ACM C-D [ 115 ] 43,380 p from AMiner, a , ACM CSS tags AAN_modified [ 5 , 49 ] 21,455 p from 312 v from NLP, 17,342 a , 113,367 c AAN_tiny [ 106 ] 2082 p (ids, titles, publication year), 8194 c , avg. 7.87 c per p , a , v Sowiport [ 28 ] u i data from Mar 2017 to Oct 2018, 0.1% click-through rate RARD_tiny [ 30 ] 800 input p from Related-Article Recommendation Dataset from Sowiport [ 13 ] CiteSeer [ 46 ] 1,100 p , 10 sets of relevant p CiteSeer_tiny [ 94 ] 400 c -pairs, 1,230 c contexts CiteSeer_medium [ 92 ] 10 p , 226 c -pairs …”

“…The rapidly increasing number of publications leads to a large quantity of possibly relevant papers [6] for more specific tasks such as finding related papers [28], finding ones to read [109] or literature search in general to inspire new directions and understand the state-of-the-art approaches [46]. Overall researchers typically spend a large amount of time on searching for relevant related work [7].…”