Application Research Of Shipping Cargo Mails Information Extraction Technology

Ocean transportation is one of the main modes of transportation in international logistics,In recent years,the main shipping business has been done via E-mail.The shipping logistics company uses the artificial way to deal with and collect the mail information.But the mail content is unstructured,the artificial processing is inefficient and it is easy to result in information identification errors,and key information omissions and so on.Meanwhile,it is asked that the business personnel have strong professional knowledge.Therefore,the extraction technology in shipping mail field has been created.The categories of shipping mail include cargo mail,tonnage mail,transaction information mail,etc..Particularly,the cargo mail that the consignor provides the basic information of the goods to be transported.Because of the large number of such mails and complexity of contents,shipping cargo mails were selected as the research object.This paper explored the extraction method of maritime cargo mail,designed and implemented the maritime cargo mail information extraction system,and the specific research contents were as follows:1.This paper proposed an extraction method for shipping cargo mails based on deep learning.This method builts the extraction model of Bi LSTM-MHA-CRF based on Bi LSTMCRF model by integrating the multi-head attention mechanism(MHA)into this method.Multi-head attention mechanism can extract relevant information from different dimensions and representation subspaces.Convolutional Neural Network(CNN)extracts character-level features of words,and combines word vectors as inputs to the BILSTM-MHA-CRF model,which effectively improves the extraction effect.2.This paper proposed an integrated deep learning and rule matching method for maritime cargo email information extraction In order to solve the problem that deep learning method is not effective in extracting some entities,regular expression rules were designed for information extraction,and a new method integrating deep learning and rule matching for maritime cargo email information extraction is proposed.By summarizing the characteristics of each entity and comparing the performance of each entity in the deep learning method and the rule matching method,this method combined the advantages of the two methods to improve the extraction effect.3.This paper built semi-automatic labeled experimental data.To solve the problem of high cost of traditional manual annotation data,CRF open source tool was adopted and four features of word,part of speech,solid boundary and suffix were considered to complete the design of feature template and predict the unlabeled corpus,so as to relieve the pressure of manual annotation and solve the problem of lack of training data.4.The validity of the proposed method was verified by experiments.Experimental results showed that the extraction capacity of Marine cargo mail based on deep learning reached 75% of F1 value,and the extraction method integrating deep learning and rule matching reached 80.3% of F1 value.The effectiveness of this method was better than that of deep learning and rule matching,which proved its effectiveness.5.This paper designed and implemented marine cargo mail information extraction system which includes four functional modules: mail management,extraction management,extraction application management and system management.The system has been tested and can be used in practice.