融合注意力机制的MacBERT一DPCNN农业文本分类模型

打开文本图片集

中图分类号：TP391 文献标识码：A 文章编号：2095-5553（2025）08-0083-07

Abstract：Aimingatthecharacteristicsof high informationdensity，ambiguous semantics and sparsefeatures inagricultural texts，this paperproposeda novel text clasificationmodel fortheagriculturalsector，basedon MacBERT（MLMas correction-BERT），deep pyramid Convolutional network（DPCNN）and Atention mechanism，named MacBERT— DPCNN—Atention （MDA）.Firstly，MacBERT was employed tocomprehensively extract contextual information，which strengthenedtherepresentationof semantic featuresof text.Subsequently，DPCNNmodel wasutlized toeffectively capturethelocal textfeatures through itsmulti-layeredconvolutional neural networkandpoolingoperations.Finaly，an Atentiomechanism was incorporated to enrich the feature representationof agricultural text sequences.The experimental results show that the precision rate of MDA model in agricultural text classification task is at least 1.04% higher，the recall rate isat least 0.95% higher and the F1 value isat least 0.14% higher compared with the other mainstream models such as BERT—DPCNN，BERT—CNN and BERT—RNN. These fidings fully confirm the efectiveness and superiority of the proposed model in addressing the issues of text classification within the agricultural domain.

Keywords：agricultural text classification；MacBERT model； DPCNN；attention； pre-trained model

0 引言

随着农业科技的迅速发展与信息化进程的加速推进，大量关键性农业信息以文本的形式快速涌现1，如苗情、情、病虫害信息等。（剩余11546字）