基于Adaboost与朴素贝叶斯的农业短文本信息分类

Agricultural Short Text Information Classification Based on Adaboost and Naive Bayes

朴素贝叶斯分类器过分依赖分类数据的质量,当待分类数据呈现复杂多元属性时,其分类的效果急剧下降,利用adaboost算法组合多个朴素贝叶斯分类器设计A_B模型。将3600份原始数据经过中文分词、句法分析、文本向量化后将A_B模型训练成一个A_B分类器。解决了分类器对于待分类数据敏感的问题,两个A_B分类器协同工作将二分类器转换为三分类器,解决了将原始农业文本信息分为农业新闻类,农业技术类,农业经济类三种类型的问题。分别利用600份标准数据与加了30%干扰信息的复杂数据测试分类器的分类效果,实验结果表明A_B分类器不仅对标准分类数据具有良好的分类效果,面对复杂多元的分类数据是仍然表现出较好的分类性能。利用不同的测试数据对A_B分类器测试发现:A_B分类器均具有良好的收敛性,其分类效果不依赖分类数据特征,具有分类效果的稳定性。

Naive Bayes classifier relies too much on the quality of classification data.When the classified data presents complex multivariate attributes,whose classification effect decreases sharply.Adaboost algorithm is used to combine multiple Naive Bayesian classifiers to design A_B model.After Chinese word segmentation,parsing and text vectorization,the A_B model is trained as an A_B classifier based the 3600 sets of original data.The problem that classifier is sensitive to data to be classified is solved.Two A_B classifiers work together to convert two two-category classifiers into one three-category classifiers,and solve the problem that the original agricultural text information is divided into three types:agricultural news,agricultural technology and agricultural economy.Using 600 sets of standard data and complex data with 30%disturbed information to test the classification effect of the classifier,the experimental results show that the A_B classifier not only has a good classification effect on the standard classification data,but also has a good classification performance to complex and multivariate classification data.Using different test data to test A_B classifier,it is found that A_B classifier has good convergence,whose classification effect does not depend on the characteristics of classification data,and has the stability of classification effect.