Research On Chinese Named Entity And Relation Extraction Algorithms And Related Applications

Chinese named entity and relation extraction technology plays an important role in many Chinese text analysis applications.Over the years,Chinese named entity and relation extraction algorithms and technologies have been extensively studied and developed rapidly.The development of pretraining language model has further promoted the improvement of its effects,and has been used in many applications such as knowledge graph construction,machine translation,and automatic question and answer,which achieved good application effects.However,many existing Chinese named entity extraction algorithms lack the use of word information and have a single tagging framework.The relation extraction algorithm has the limitation of not making full use of entity and context information.In particular,in the field of policy text analysis and application,there is still a lack of efficient and accurate entity and relation extraction algorithms and related application technologies for policy text.Therefore,this paper has carried out the research of Chinese named entity and relation extraction algorithm,and proposed a Chinese named entity extraction algorithm based on multi-tagging scheme and fusion of dictionary and pinyin features,and a relation extraction algorithm based on entity segmentation and dependency syntax features.Then,verify the application of the algorithm proposed in this paper on policy services.The research content and contribution points of this paper mainly include:(1)The research proposes a Chinese named entity extraction algorithm based on a multi-tagging scheme and fusion of lexicon and pinyin features.By fusing the features of lexicon and pinyin,the impact of word segmentation errors on the accuracy of entity recognition is alleviated,and word information is explicitly introduced to enhance the model's recognition of entity.In addition,by jointly learning of sequence tagging scheme and pointer tagging scheme improves the accuracy of Chinese named entity recognition by taking advantage of the advantages of different tagging scheme.(2)The research proposes a relation extraction algorithm based on entity segmentation and dependency syntax features.By inserting segmentation markers at the beginning and end of the entity,and introducing relative position information to the entities for each token,then the entity information and entity position in the sentence are highlighted.In addition,carry out sequence modeling of dependency paths between entities,precisely capture the associations and dependencies between entities,and improve the accuracy of relation extraction.(3)Facing the demand for policy services,research and propose a rule-based policy requirement logic relation analysis and policy matching degree calculation algorithm.By manually establishing the key words of the analysis rules,constructing the logical analysis tree of the policy requirements,obtaining the corresponding logical expressions,realizing the analysis of the logical relation between the policy requirements,and clarifying whether the policy requirements are disjunctive or conjunctive relation.Then design a policy matching degree calculation algorithm based on conjunctive normal form to realize the calculation of the matching degree between policies and enterprises.Finally,based on the algorithm proposed in this paper,the policy service application verification was carried out,and a set of policy service application system including policy parsing and policy matching was designed and implemented.At present,the system has been successfully implemented in a district government enterprise service in Nanjing,Jiangsu Province.The platform has achieved good application results.