Construction And Application Of Knowledge Graph Of Architectural Literature Based On Deep Learning

With China’s urbanization process continuing to advance,Huizhou-style architectural require renovation and reconstruction work.Throughout the construction process,architectural researchers must consider how to better preserve and carry on the fine architectural style of ancient buildings.In order to create construction plans,they need to refer to a large volume of relevant literature.However,the wealth of architectural literature has a dispersed knowledge distribution,weak correlations,and contains a great deal of invalid information.Traditional search platforms’ databases are unable to establish deep connections between the unstructured contents in the literature,and the literature search process will excessively consume the time and energy of researchers.To address the issues of traditional retrieval platforms,this paper aims to improve the efficiency of literature retrieval by constructing a knowledge map of architectural literature as the knowledge base of the retrieval platform.By utilizing the powerful knowledge interconnection capability of the knowledge map,it enhances the correlation between knowledge in literature.The specific work involves named entity recognition,relationship extraction,knowledge graph construction,and the development of a literature visualization retrieval platform.The automatic knowledge graph construction model designed in this paper is integrated into a literature retrieval platform to enable architectural researchers to efficiently access information from the literature.The main work of this paper is as follows:(1)Design a model for recognizing architectural named entities by fusing dual features of word granularity and radical granularity.The model first uses the pre-trained language model XLNet to obtain a vector representation of architectural literature,and then uses a bi-directional long and short term memory network(Bi LSTM)to extract word-grained features.Secondly,in combination with the radical features of architectural vocabulary,it uses an iterative expanded convolutional neural network(IDCNN)to extract the feature at the particle size,and uses a Co-Attention Network to integrate word-grained and radical-grained feature vectors into a<word-radical> dual feature vector.Finally,the model uses Conditional Random Fields(CRF)as the label decoding optimizer to output recognition results.Tested on the emblematic architecture literature dataset,the model achieves an accuracy,recall and F1 values of 93.29%,92.68%and 92.98%,respectively.(2)Design a model for building entity relationship extraction by fusing dual features at the word-level and sentence-level.The model takes word vectors as input,maintains the sentence’s integrity through bi-directional gated recurrent units(Bi GRU),and incorporates a Dual Attention mechanism to extract both word-level and sentence-level text features.By continuously adjusting the weight proportion between valid and invalid information in the text sequence,the interference caused by noisy data and loss of valid information can be minimized.On the emblematic architecture literature dataset,experimental results demonstrate that the model’s extraction efficiency is significantly improved after adding sentence-level features,with accuracy,recall,and F1 values of 75.55%,75.39%,and 75.46%,respectively.(3)Building a knowledge graph for Huizhou-style architectural literature,and a bottom-up approach is employed for its construction.This approach consists of four levels: the data source level,which uses crawler technology to obtain Huizhou-style architecture literature;the data pre-processing level,which employs YEDDA and Brat annotation tools to complete data annotation;the model training level,which trains the entity recognition and relationship extraction models designed in this paper;and the knowledge graph construction level,which involves storing the obtained entity and relationship triples in the Neo4 j graphical database.Data pre-processing is crucial as it allows the model in the training layer to perform supervised learning on the annotated data,effectively improving the accuracy and generalization ability of the model.Additionally,the use of the Neo4 j graph database facilitates efficient knowledge graph construction and querying by storing entity and relational triples.(4)Development of a visual search platform for architectural literature based on knowledge graphs.To meet the needs of architectural researchers for the integration of literature knowledge query and construction,this paper combines architectural literature with knowledge graphs,and uses the Flask framework and the entity identification and relationship extraction models designed in this paper to develop a visual retrieval platform for architectural literature to achieve automatic architectural knowledge extraction and literature visualization search functions.Figure [46] table [23] reference [65]...