Research On Importance Evaluation Method Of Scientific Literature Based On Multiple Fusion Information

With the rapid growth of information in the Internet,researchers obtain information through literature retrieval platform.The number of researchers and scientific literature are increasing dramatically,how to make researchers find the most relevant literature in scientific literature database,we need to get the importance of scientific literature.The subject of this paper is to evaluate the importance of scientific literature.The current evaluation methods of the importance of scientific literature are: 1)evaluate scientific literature according to citation counts;2)evaluate scientific literature based on PageRank algorithm;3)evaluate scientific literature based on HITS algorithm.These algorithms have some limitations in the evaluation of the importance of scientific literature: depending directly on the citation counts can be intuitive to see the results,but there is relatively insufficient in the actual application.Scientific literature published earlier can get high citation counts,but recently published scientific literature on behalf of the latest research results can get few citation counts impossibly because of being published in a short time.Evaluation of scientific literature using PageRank algorithm is based on the whole network,the importance calculation not only rely on citation counts,but also depend on the importance of scientific literature.Evaluation of scientific literature based on HITS algorithm has the problem of not using other related attributes.Our paper proposes an effective model named FusionRank focusing on the problem existing in the scientific literature evaluation.FusionRank uses journal node,journal-year composite node,scientific literature node and author node to construct multiple information fusion networks.We design and implement the FusionRank model.The main research contents include:1)Research on the factors affecting the importance of scientific literature: each scientific literature has its basic properties,and may have its own unique properties.In our evaluation model,it is essential to make full use of related attributes and analyze their influence on the scientific literature.Factors mainly refer to the citation counts,the impact factor of journal,the importance of author and the year of publication.2)Study multiple information fusion networks: any network structure is composed of the connection between the nodes and the nodes.On the one hand,the kind of nodes is various,first the network must contain the nodes to be evaluated,but also contains the other nodes associated with the evaluation of other nodes.On the hand,the association between the nodes and the nodes is also common in the construction of multiple information fusion networks,the relationship associated with the problem should be contained into network,and the others can be ignored.Constructing fusion network in this way can get low time complexity and high operating efficiency.3)Construct the evaluation model using multiple relations in dataset based on multiple fusion information,design and implement it.Our experimental evaluation results on real DBLP dataset show that our proposed FusionRank has good performance in validity and feasibility compared to PageRank.