Design And Implementation Of Scientific Literature Retrieval System Based On Graph Structure

Scientific literature retrieval based on vertical domain search can enhance the retrieval performance and support scientific research activities by matching the input retrieval keywords and the structural semantics of the literature.Semi-supervised literature and retrieval graph structure semantic model are used to enhance the accuracy of literature retrieval.On this basis,off-line and online pipelining system architecture is designed,and a variety of structural semantic models is integrated and can be configurable.The specific work includes:1.Retrieval requirements analysis,offline representation learning and online retrieval pipeline configurable architecture design: analysis of the key functions of literature map model and retrieval keyword hidden structure graph model,such as feature extraction,index,system management,database access.A pipelined system structure from presentation learning to retrieval,which separates off-line representation learning from online retrieval is proposed,and a configurable and pluggable modular semantic fusion system architecture for a variety of representation learning algorithms is proposed.2.Semi-supervised literature and retrieval graph structure semantic model: the co-occurrence relationship of words and words,the inclusion relationship of words and literature,and the literature citation relationship constitute heterogeneous graphs,and the structural semantic representation of words and literature is extracted by various ways.The retrieval keyword set is treated as a short literature,and the word co-occurrence relationship in the literature graph structure is generalized to the retrieval keyword set,and the hidden structure graph of the retrieval keywords is established.The semantic of the literature graph structure is introduced to extract the structural semantics of the retrieval keyword set.3.Multi methods semantic fusion retrieval: according to the characteristics of semantic representation obtained by various methods,semantic fusion is carried out.Through the structural semantics of literature graph and retrieval keyword set after fusion,semantic matching between literature and retrieval keyword set is established,and configurable retrieval service is provided.Finally,through the prototype system implementation and testing,the main functions of the scientific literature retrieval system are realized.On the basis of full experience evaluation of the performance of the key modules of the system,the integration test of each functional module of the system is carried out.The test results show that the system can meet the user's core function and performance requirements.