Research Of Named Entity Recognition Of Apple Diseases And Pests Based On Word Augmentation

In the long history of apple cultivation in China,the problem of diseases and pests has always been regarded as the most common factor restricting the development of the apple industry.When apple planting is attacked by diseases and pests,most apple growers are unable to prescribe the right medicine for the diseases and pests due to lack of professional knowledge,which may delay the best control time of diseases and pests and make the situation worse.Therefore,it is of great practical significance to extract effective information from the professional literature in the field of apple diseases and pests to help apple growers to better solve the problems of diseases and pests.Among them,named entity recognition is particularly important as the most basic key step in the information extraction task.At present,the research on named entity recognition in the field of apple diseases and pests is still in the early stage of development,and there is a lack of relatively mature Chinese named entity recognition datasets.In addition,the corpus in the field of apple diseases and pests is highly specialized,including many domain terms and proper nouns,and has the characteristics of various entities categories,entities with aliases or abbreviations,and the difficulty of identifying rare entities.In response to the above problems,this paper conducts a research on named entity recognition for the texts in the field of apple diseases and pests.The main research contents are as follows:(1)In view of the lack of mature named entity recognition datasets in the field of apple diseases and pests,this paper screened professional literature in the field of apple disease and pest control,and preprocessed the texts such as deduplication and data format conversion,and finally constructed a Chinese apple disease and pest NER dataset named ApdCNER that fill a gap in current research in this field.This dataset contains a total of 21 entity categories,5574 samples and 11876 entities,which lays the foundation for this research and subsequent research.(2)According to the characteristics of apple disease and pest domain corpus,this paper proposes a named entity recognition model APD-CA for this domain corpus.This model uses the character-based BiLSTM-CRF model as the baseline model.Aiming at the problem of the poor recognition of rare entities and aliases,we incorporate dictionaries into the model to provide more semantic information and entity boundary information and design a vocabulary fusion method that enables the model to process multiple sentences at the same time.In view of the problem of having less data for some entity categories,we supplement the sentence semantics by integrating similar words to compensate for the lack of data.(3)In this paper,comparative experiments are designed from multiple perspectives to study and analyze the proposed apple disease and pest named entity recognition model APDCA.The experimental results show that the precision,recall,and F1-score of the APD-CA model based on ApdCNER are 92.29%,91.99%,and 92.14%,respectively,which are 2.95%,2.13%,and 2.54% higher than the baseline model BiLSTM-CRF.Compared with the other 4SOTA(state-of-the-art)models,they are also improved.The improvement verifies that the proposed model in this paper has performance advantages in named entity recognition task in the field of apple diseases and pests.Other experimental results also prove that this model has efficiency advantages and certain generalization advantages.The research in this paper effectively improves the performance of the named entity recognition task in the field of apple diseases and pests.It provides the underlying technical support for downstream research work such as knowledge graph construction,intelligent question answering system,and intelligent semantic search in the field of apple diseases and pests,and helps apple growers to find the corresponding disease and pest control methods more efficiently and conveniently.