Knowledge Graph Construction For Technology Patents And Design And Implementation Of Service Components

As the most effective carrier of technological information,technology patents encompass the majority of the latest global technology intelligence.However,technology patents resources have not been fully utilized.The text of technology patents is characterized by difficulties in obtaining,heterogeneity from multiple sources,and diversity in storage methods,make it difficult to establish the relationships among patent entities.These characteristics also make it more difficult to mine statistical patent information and potential patent entity relationships,reducing the efficiency of subsequent search and reasoning work.Therefore,the construction of a technological patent knowledge graph and the design and implementation of service components are of great significance for scientific research.It not only helps researchers quickly obtain information but also promotes the development of scientific research.The primary contributions of this thesis are as follows:(1)We proposed a technology patent entity relation extraction method based on feature reasoning(TPEFR)to address the problem of relation overlap and bias propagation in the process of entity relation extraction in the field of technology patents.TPEFR integrates entity recognition and entity relation extraction tasks,and extracts entity relation triplets from multi-source scientific and technological patents.A table feature is created for each predefined relation,which is combined with a feature related to the subject and object.TPEFR based on Transformer is adopted to reason the entity relation features,and historical features are integrated to better reveal the differences between relations and entity pairs.The experiment shows that TPEFR improves the precision and recall of triplet extraction.(2)We proposed a technology patent entity alignment algorithm based on graph convolutional network(GEATP),which realizes the knowledge fusion of technology patent knowledge graph.By combining graph structure representation with attribute semantic representation,and learning the approximate semantic representation of potential alignment entities,GEATP achieves unified modeling of multi-source and heterogeneous technology patent knowledge graphs.GEATP uses graph convolutional networks and pre-trained models to represent the structure and entity attribute information of the knowledge graph,embeds graph structure information and text semantic information into a unified vector space,and implements entity alignment based on similarity sorting.The experiment shows that GEATP improves the accuracy of technology patent entity alignment and constructs a more complete technology patent knowledge graph.(3)We proposed a knowledge graph-based technology patent knowledge reasoning method(KGRTP).KGRTP is based on the knowledge representation method TransE,which translates the technology patent knowledge graph into vector representation and preserves the translation characteristics of the graph through the translation model.A multi-head attention mechanism is introduced to learn different relationships and weights between technology entities.KGRTP obtains high-order neighbor information of the technology patent knowledge graph and extracts deep semantic information based on convolutional neural networks to adapt to different reasoning tasks.The experiment shows that KGRTP can improve the efficiency and accuracy of completing and reasoning tasks in technology patent knowledge graph.(4)We designed and implemented the service component of knowledge graph of technology patents.The service component provides functionalities such as knowledge retrieval,patent information statistics,and relationship visualization.The service component integrates the functionalities of entity relationship extraction,entity alignment,and knowledge inference.The service component is validated and integrated into the technology patent knowledge graph.Additionally,a mobile app is developed based on the technology patent knowledge graph service component to provide users with convenient and efficient technology patent knowledge services.