Question Classification System For Agriculture

Traditional agricultural technology services mainly rely on expert visits,training,and telephone communication,which have the advantages of targeted and real-time services.However,the shortage and uneven distribution of experts have resulted in problems in cost,timeliness,and resource allocation for agricultural technology services.With the development of information technology,farmers are gradually turning to online methods to seek agricultural technology services.However,there is a massive amount of redundant data on the internet,which interferes with farmers’ information queries.Agricultural intelligent question-answering systems can automatically answer users’ questions,effectively improve the efficiency of farmers’ queries,and provide support for agricultural production.The primary task in the question-answering system is to classify the questions,and the classification performance plays a crucial role in the subsequent answer extraction.In view of this,this thesis collects and organizes agricultural Q&A texts and establishes an agricultural Q&A corpus.Machine learning,deep learning,and pretrained models are used to build agricultural question classification models,and web technology is used to implement an agricultural question classification system.The specific work of this thesis is as follows:(1)In response to the lack of agricultural Q&A corpus,a crawler program is used to automatically crawl relevant texts,and the data is organized and saved in a My SQL database to construct an agricultural Q&A corpus.A total of 100,284 Q&A text data in six categories are collected and analyzed and visualized.(2)In response to the characteristics of short text length,sparse features,and nonstandard expression,pre-trained language models and deep learning models are used to construct classification models.Firstly,based on the BERT model and the improved BERT-WWM for Chinese tasks,fine-tuning is conducted,and the experimental results show good fitting ability.Secondly,using RNN models for classification,a semanticenhanced question classification method based on the Word2 Vec model can be proposed with the help of resources in the agricultural field to expand the semantic information of the questions.Multiple RNN models are used for experiments,and the Bi-LSTM-Att model achieves the best performance.The two models constructed in this thesis have the characteristics of high recognition accuracy and fast recognition speed.(3)Based on the above Q&A texts and the constructed classification models,a B/S-based classification system is designed and implemented.Users can use this system without installation,which includes functions such as login and registration,question classification,classification model selection,result feedback,and history record viewing.It realizes the classification of user questions,provides help for subsequent answer extraction,and user feedback on the classification results can also be used to improve the classification performance of the model.