| With the rapid growth of the number of papers,academic researchers are facing the problem of information overload.Paper recommendation systems can help scholars find the required literature from the huge amount of literature data.Currently,hybrid recommendation methods that combine paper content and network structure have become a new research trend.However,most of the existing hybrid recommendation methods only consider textual information and citation relationships.Lack the integration of information about academic entities and do not distinguish the importance of cited literature.Which is not conducive to producing more effective recommendation results.Therefore,this thesis combines the content of papers and academic networks to first find papers for scholars with similar content to their historical interests.Then constructs heterogeneous academic networks using five academic entities,measures the citation intensity among papers to weight the citation networks,and finds more recommendable papers for scholars through the network structure.The main research work of this thesis is as follows:(1)A paper recommendation algorithm based on content similarity is proposed.The literature data is obtained and analyzed through the AMiner academic platform.The abstract is selected to represent the content information of a paper,and the original literature data is pre-processed to construct the dataset used for the experiment.The processed abstract data are passed through the paragraph vector model to obtain the paper feature representation.The scholar feature representation is obtained based on the historical interest of scholars.The cosine similarity between scholars and papers is calculated and ranked to generate a preliminary recommendation list.Expand scholar interests using papers with similar content,and then build a scholar-paper interest network.Provide a basis for subsequent research work.(2)A paper recommendation Algorithm based on content and academic network is proposed.Based on the scholar-paper interest network,citation relationships among papers and academic entities such as paper authors,paper publication sites,and paper research fields are added to construct a heterogeneous academic network,and meta-path analysis is performed on the network.The citation intensity is defined using a combination of three metrics: literature coupling and literature co-citation similarity,scholar interest similarity and paper label similarity,and weighted for the citation network.The transfer probability matrix is calculated based on the relationship matrix of the network.The recommendation scores of candidate papers for the target scholars are calculated using the biased random walk with restart algorithm in link prediction,and the recommendation list is generated after ranking.Experiments are conducted on the AMiner dataset,and the experimental results show that the algorithm has a good performance in terms of Precision,Recall and F1 value.(3)A paper recommendation system is designed and developed based on the proposed algorithm.It provides personalized paper recommendations for scholars proactively according to their different interest needs.Scholars can log into the system to get a daily list of paper recommendations,search for desired academic papers,learn about popular search terms in the field,read the content of papers in detail,collect papers of interest,and view recently viewed and all collected papers.The system has been tested for its usefulness. |
