Research On Food Safety Event Named Entity Recognition Based On Deep Learning

In recent years,endless food safety issues have aroused great attention from all walks of life.For example," Hogwash oil ","clenbuterol" and "zombie meat" are food safety issues that have been lurking around people for a long time.At this stage,named entity recognition is still in the preliminary exploration stage in the field of food safety.Named entity recognition for multi-class food safety is even rarer.The text data in the food safety news is the key to reflect the food safety issues,whether it is the extraction of information on the food safety news,the construction of the question and answer system in the field of food safety,or the construction of the visualization of the knowledge map in the field of food safety.The task of identifying named entities for food safety incidents is the first and most critical step in solving the above problems.Based on this,this article decided to carry out research on the realization of automated named entity recognition methods in the field of food safety.The main contents are as follows:(1)Organize and build a food safety incident corpus.This thesis crawled 2,321 news reports from the China Food Safety Network in 8 areas including product substandard,counterfeit,illegal,over-standard,illegal,prohibited,and sentenced to fines caused by food problems in2019.Word Food Safety Corpus.(2)Construction of a dictionary of proper terms in the field of food safety.In this paper,we construct a special dictionary of food related terms issued by the relevant national agencies.The special dictionary includes 1308 special terms,including the name catalog of food safety related agencies,food classification catalog,food additive use standards and other aspects.(3)Data preprocessing.This article studies the difficulty of Chinese word segmentation in natural language processing and the labeling system for named entity recognition.This article uses a dictionary composed of proper nouns in the food field issued by relevant national agencies and jieba word segmentation technology to combine food safety event corpora.The data in the Chinese word segmentation and part-of-speech tagging are used as preliminary data preprocessing.Based on the analysis and research of the data in the corpus,seven types of named entities are proposed.According to the type of classification,the preprocessed data of the preliminary data is further refined for more accurate data processing tasks.Manual BIO sequence labeling.With the intervention of the dictionary in the field of food safety,the error rate of manual BIO labeling will also be greatly reduced.Sort and summarize the labeled data to obtain the data labeled with the food safety event sequence required for training the model.(4)A model for named entity recognition in the field of food safety based on deep learning methods is proposed.This article uses the BILSTMCRF model for the first time in the multi-class named entity recognition of events in the field of food safety.On this basis,in order to increase the model's attention to the seven classification entities,this paper proposes to add the attention mechanism to the BILSTM-CRF model,so as to increase the model to the seven classification entities of food safety incident Weight value of multi-class entities to achieve higher accuracy.The experimental results show that the ATTENTION-BILSTM-CRF model is better for identifying the seven types of named entities.Precision has improved by an average of 4.86%,recall has increased by an average of 4.957%,and F1-measure has improved by an average of 5.19.For the overall named entity recognition,Accuracy is up to 99.16 %,F1-measure is 3.08 higher than the BILSTM-CRF model,reaching 94.51.In this thesis,two models are used to automatically identify the seven categories of named entities for food safety news event corpora,and it also proves that the trained ATTENTIONBILSTM-CRF model can perform the task better.