Research On Knowledge Fusion Technology Of Public Security Knowledge Graph

As a highly interconnected semantic network,knowledge graph plays an important role in intelligent services.However,complex semantic relationships pose a huge challenge to knowledge fusion,resulting in lower accuracy in entity matching calculations.To this end,this thesis studies the knowledge embedding and entity alignment techniques in knowledge fusion,strengthens the semantic features of entities and relationships through attention mechanism,and improves the accuracy of fusion alignment through joint type matching in entity matching calculation.The main work and contributions are as follows:(1)A knowledge embedding model based on translation and attention mechanism is proposed.The traditional translation model only models the internal structural characteristics of the triples,resulting in lower accuracy in link prediction.In this thesis,the global attention mechanism is used to assign attribute node features to the entity representation,and the self-attention mechanism is used to extract the text feature from the tag word sequence for relational representation.It solves the disadvantage of insufficient semantic features of entity and relational representation.(2)A cross-knowledge graph entity alignment model for joint type matching is proposed.Existing methods based on knowledge embedding only perform matching calculation based on the similarity of the entity vectors,resulting in low alignment accuracy of the entities.In this thesis,we use the entity type feature to attach the type matching constraint to the alignment model to enhance the selection condition,and use the iterative entity alignment method to overcome the problem of less label data.(3)A system for constructing public security chinese knowledge graph has been realized.Most of the current knowledge graphs belong to the general category and cannot meet the needs of specific fields.This thesis designs and completes the modules of domain data acquisition,knowledge extraction,knowledge storage and knowledge fusion,and forming a professional domain-specific knowledge graph,which can provide technical support for intelligent service applications.