Research On Information Extraction Technology Of Text Intelligence Based On Deep Learning

With the popularization of computer technology and the rapid development of Internet technology,in order to meet the challenge of data explosion in the information age,information extraction technology emerged and developed rapidly.In battle command,intelligence data has the characteristics of huge volume and complex structure.How to deal with large-scale intelligence data online,and then form small-scale and easy to analyze summary information,improve the efficiency of intelligence processing,break through the limitations of traditional data technology,has become an important practical problem in battle command.In order to solve these problems,this paper will focus on the technology of text information extraction.Because military information belongs to state secrets,there is no open military information data set that can be used for in-depth learning at present.Therefore,this paper collects a large number of open military news with intelligence characteristics from the Internet according to the real intelligence characteristics to form a simulation intelligence data set.According to the attention content of such intelligence as the Department’s activity,a set of text intelligence sequence based on the BIO annotation set is proposed.Column annotation system.After the completion of data collection,because there are significant differences among the entities concerned by different categories,we need to extract information from intelligence classification.Therefore,this paper proposes a network structure combining CNN and Bi LSTM,introduces Attention Mechanism to enhance the attention of text keywords,establishes att cbilstm model,and combines the advantages of CNN,Bi LSTM and Attention Mechanism in text processing.Experiments show that the model can complete the task of packet classification better,and the validity of the model is proved by comparing experiments with several classical models with different structures.n the task of named entity recognition,we first try to introduce the current excellent Bi LSTM-CRF model into the named entity recognition of force text intelligence.We find that the model performs well in three kinds of tags: prediction time,country(region)and number of personnel,in the prediction of active position,and in the prediction of behavior subject.At the same time,we also try to introduce the Attention Mechanism into the model.The experiment shows that the F1 value of each tag increases by 0.4% on average after the introduction of the Attention Mechanism,which proves the effectiveness of the introduction of the attention mechanism in packet named entity recognition.