Design And Implementation Of Financial Nested Named Entity Recognition System

Data is the driving force for social and economic development in the information age.The financial industry relies on data-driven and is also the lifeblood of social and economic development,so its security has attracted much attention naturally.Practice has proved that only a combination of macro policies and emerging technologies can effectively guarantee the all-round security of financial data.As an emerging artificial intelligence technology,knowledge graph can structure data and extract specific content.Applying knowledge graph to the financial field can quickly extract the key information of financial data,which is convenient for subsequent targeted protection of data.And as one of the basic tasks of constructing knowledge graph,named entity recognition is crucial to the construction of financial knowledge graph.The financial field is different from the general field,in which the entities are generally long and the nested form of entities is more complex.In particular,there are many nested named entities in the names of financial institutions and financial products,if these entities are ignored,a lot of semantic information will be lost.Nested named entity recognition aims to identify all entities in the text completely,which is conducive to obtaining richer entity information and deeper semantic information,and improving the quality and quantity of entity recognition.Now,the nested named entity recognition is mainly carried out around the general field.Consequently,this paper mainly focuses on the following investigations based on the traits of intricate and lengthy nested entities in the financial realm:1.A financial dataset containing nested named entities is constructed to test and verify the performance of subsequent models and systems conveniently.We collected 12000 texts in the financial field,and identified five types of entities and several types of nested entities,then completed the construction of financial nested named entity datasets after manual screening,cleaning,inspection,and labeling.The dataset contains 8000 texts,of which the labeled entities are relatively complete,with a large number,strong professionalism,high nesting rate,which has strong practicability and domain characteristics.It can be used for simple named entity identification and nested named entity identification,and additional annotation can also use for entity relationship extraction.2.A hierarchical recognition model for nested named entities based on semantic enhancement is proposed.A semantic enhancement algorithm devised in this paper,is designed to reduce data sparsity and enhance the recognition effect.At the same time,the structure of hierarchical recognition of entities of different lengths is adopted to avoid the error propagation problem of the cascaded nested named entity recognition model and improve the accuracy of recognition.The results in the experiment prove the excellent performance and strong generalization ability of the model.3.A nested named entity recognition system for the financial field is designed and implemented.In this paper,an end-to-end nested named entity recognition system is designed,which not only supports the verification of the performance of Nest NER algorithms,but also allows users to choose the data set to be identified,then view the recognition performance on the data set,and show the results of nested named entity recognition to realize the visualization of the results.The system in this paper has excellent performance,strong generalization and broad application prospects.