The Research Of Named Entity Recognition In Agricultural Field

Named entity recognition(NER)seeks to locate and classify named entity mentions in unstructured text into pre-defined categories,which a crucial basic task in natural language processing(NLP)and has been applied in many high-level applications of natural language processing.In the era of agricultural information,there are a lot of unstructured agricultural texts stored in the Internet.It is the key to realize intelligent agriculture that how to transform these disordered agricultural texts into high-quality and structured agricultural information and knowledge.Therefore,agricultural named entity recognition is a very meaningful research topic.This thesis mainly studies how to use the neural network model to recognition nine kinds of named entities which related to the agricultural,and then analyzes the advantages and disadvantages of the proposed models from the aspects of recognition effect,training time,decoding time and sucking up video memory.The main work and innovation points of this thesis are summarized as follows:(1)In this thesis,we chose the current popular neural network model which is(bi-directional Long Term Memory)LSTM network combined with Conditional Random Field(CRF)as the benchmark model.Then optimize and improve it,and propose a stack BiLSTM model based on dense connections,DC-BiLSTM.And we introduce the attention mechanism to pay attention to the entity fragments in the text,increase the difference between the entity part features and the non-entity part features,so that the entity is more'prominent' relative to the non-entity,which is convenient for CRF prediction of each character label.(2)We propose a named entity recognition model based on deep attention mechanism--Deep Attention,which combines BiLSTM with multi-head attention mechanism.BiLSTM is used to obtain the association between contexts,so that the feature vector has time series;the multi-head attention mechanism is used to learn the feature information of different subspaces,and the semantics of words,syntax and different levels are mapped into multiple multidimensional vector spaces.The experimental results show that the Deep Attention model achieves optimal results in many aspects such as recognition effect,training time,decoding time and memory usage.(3)We gather three kinds of NER model which use in the experiments in this thesis,and then design a NER system.We introduce the training module,test module and result display.The system automatically recognizes the text input by the user,extracts the named entity related to agriculture,and finally displays the recognition result through the webpage.