Research On Ranking Scientific Publications Based On Citation Graph

Academic paper ranking algorithm is a technique to rank a large number of academicpapers, so that it is possible to fetch the right papers in vast amounts of scientific literatureand recommend to different grade researchers with right papers. All of above are required tostudy the technology. In this paper, base on the analysis of the current literature rankingalgorithm, we proposed two academic literature ranking algorithms whose ranking results arebetter and more in line with the practical needs.In the case of applying in the field of literature, we found that it is impossible to use thedata sets involving all papers, which would cause that the reference network diagram isincomplete and result in many abnormal phenomenon. For example: a paper is referenced bymany other papers, but only one reference paper is appear in the reference network diagram,which made the weight of the source paper be very small. As described above, though thepaper is perfect, its weight would be small because most of the reference papers are absent inreference network diagram. If the paper reference to other papers, it would make such papers’weight be small too. So as to overcome this problem, we propose a ranking algorithm basedon the external links called ELRank ranking algorithm and the extended N-ELRank rankingalgorithm to fix this problem. The experiments show that, the correlation between ELRankand CitationCount is much higher than that between PageRank and CitationCount. Theproposed algorithm already achieve the desired effect, the weight of perfect paper would notbe small as only one reference paper is present in data set, so that lead to rank mistake.Most of the current literature ranking algorithm didn’t comprehensive consider theinfluence of time and attention but calculate the weights of the entire literature of the timeperiod, which makes the old article get a lot of references by the accumulation of time andeasily result in high weights to the old posts, reducing the weights of new article. In order toimprove this problem, make the new articles and old articles distinction. In this paper, we addthe paper type, attention factor and time slice into the proposed algorithm called TSRank. Theexperiments show that, the top100perfect papers result from TSRank are published in lastfive years while the result papers from other algorithms are published in last stage of1990s.TSRank is more in line with people who like the new high-quality literature than otheralgorithms.