Construction Of Intelligent Question-and-answer System For Agricultural Mechanization

With the exponential growth of network data content,it is more difficult to obtain the specified information in the huge and complex Internet data.Compared with the traditional search engine,the user returns the relevant pages by keyword matching the sentences entered by the user,and then the user filters the data he needs from the search pages.The question answering system for specific areas has the advantages of efficient retrieval and strong pertinence.Knowledge graph is used to connect the relationship between various entities.Its knowledge organization form and graphical relationship display mode have attracted the attention of many industries.Its rapid development also promotes the application of question answering system based on limited fields in all walks of life.At present,in the field of agricultural mechanization,farmers mainly encounter two major types of problems in the process of production,namely technical problems and policy problems.Technical problems are closely related to the crops and agricultural machinery used by farmers.In terms of policy,farmers are most concerned about agricultural machinery purchase subsidy policy.In this paper,the rape machine sowing technology is selected in the technical aspect,the agricultural machinery purchase subsidy policy is selected in the policy aspect,and the intelligent question answering system is constructed.The system has the advantages of high efficiency,convenience and pertinence compared with the traditional farmers’ information collection methods.The main research in this article is as follows:(1)Knowledge map construction of rapeseed mechanized sowing and agricultural machinery purchase subsidy.At present,in the field of agriculture,there is no public knowledge map of rape mechanized sowing and agricultural machinery purchase subsidy.In this paper,according to the high search frequency in the collection and collation network and the related issues of rape mechanized sowing and agricultural machinery purchase subsidies that farmers are more concerned about in the actual operation,the relevant literature,websites,indigenous crops and statistical data released by government departments are targetedly collated as the data source of rape mechanized sowing,and the relevant data of agricultural machinery purchase of farmers in Hubei Province published by government departments and the relevant policy documents of agricultural machinery purchase subsidies issued by government departments of various provinces and cities in China are used as the data source of agricultural machinery purchase subsidies.By sorting out the data from different sources,dividing entities,and designing the relationship between entities,the obtained data are stored in the form of triples,and the knowledge map of rape mechanized sowing and agricultural machinery purchase subsidy that can be used for automatic question and answer is constructed.(2)Research on Question Classification Algorithm Based on Rapeseed Mechanized Seeding and Agricultural Machinery Purchase Subsidy.Different classification methods have their own characteristics in the actual classification tasks.According to the specific classification tasks,the optimal classification algorithm matching them should be selected to ensure the efficiency,accuracy and complexity of classification.In the template of rape mechanized sowing problem constructed in this paper,the characteristics between questions and questions are obvious.The calculation logic of naive Bayesian algorithm is relatively simple and there is no complex matrix calculation.In view of the obvious advantages of such question classification,the naive Bayesian algorithm is selected to classify the user statements of rape mechanized sowing constructed in this paper.Then,the TF-IDF algorithm is used to extract the characteristic words of rape mechanized sowing,and convert them into text feature vectors,and the generated text feature vectors are input into the naive Bayesian classifier.After training,the classification model of rape mechanized sowing questions based on naive Bayesian algorithm is obtained,which is used to classify the questions related to rape mechanized sowing.In the query template of agricultural machinery purchase subsidy problem designed in this paper,there are few correlation characteristics between question sentences,and it is necessary to use deep learning technology to deeply analyze the statements of agricultural machinery purchase subsidy.In addition,in view of the limitation that Word2 vec word vector model cannot accurately express the meaning of the same word in different statements,BERT algorithm can obtain word vector according to different semantic environments and the characteristics of current semantic information.Therefore,in the classification process of agricultural machinery purchase subsidy questions constructed in this paper,BERT is used to train word vector,and agricultural machinery purchase is based on BERT algorithm.(3)Research on entity recognition algorithm based on Bi LSTM-CRF algorithm.Compared with LSTM algorithm,Bi LSTM algorithm increases the calculation process from the reverse direction of the sentence,which can be fully used to the following information of the sequence.Finally,the calculated positive and negative direction values are output to the output layer,so that all information in the sequence is bidirectional.CRF can separate the correlation of Bi LSTM output layer and predict tags under the premise of context correlation,which is complementary to entity recognition tasks.In this paper,LSTM,Bi LSTM and Bi LSTM-CRF entity recognition algorithm models are used to compare the data sets of rape mechanized sowing and agricultural machinery purchase subsidy.The average F1 values of these three models for rape mechanized sowing are87.18 %,89.35 % and 91.91 %,respectively.The average F1 values of agricultural machinery purchase subsidy are 87.69 %,90.40 % and 92.31 %,respectively.The results show that compared with LSTM and Bi LSTM entity recognition models,Bi LSTM-CRF entity recognition algorithm model has obvious advantages.Then,based on the Bi LSTMCRF entity recognition algorithm model,the influence of Word2 vec and BERT word vector algorithm on entity recognition algorithm is verified.The results show that the combination of the Bi LSTM-CRF entity recognition algorithm model and BERT word vector model can achieve better results.(4)The design and implementation of question answering system for rapeseed mechanized sowing and agricultural machinery purchase subsidy.In this paper,the Naive Bayes algorithm combined with Bi LSTM-CRF method is used to construct the rape mechanized sowing question answering system,BERT algorithm combined with Bi LSTMCRF method is used to construct the agricultural machinery purchase subsidy question answering system,and the voice input function is designed,which performs well in the practical application process.It can accurately answer the questions related to the user ’ s rape mechanized sowing and agricultural machinery purchase subsidy,and meet the user ’s demand for the information acquisition of rape mechanized sowing and agricultural machinery purchase subsidy.Finally,the front-end visual interactive page of rapeseed mechanized sowing and agricultural machinery purchase subsidy question answering system is displayed.