分别使用CountVectorizer与TfidfVectorizer， 并且去掉停用词的条件下，对文本特征进行量化的朴素贝叶斯分类性能测试

from sklearn.datasets import fetch_20newsgroups
news = fetch_20newsgroups()

from sklearn.model_selection import train_test_split
x_train, x_test, y_train, y_test = train_test_split(news.data, news.target, test_size=0.25, random_state=33)

from sklearn.feature_extraction.text import CountVectorizer
from sklearn.feature_extraction.text import TfidfVectorizer

count_filter_vec, tfidf_filter_vec = CountVectorizer(analyzer='word', stop_words='english'), TfidfVectorizer(analyzer='word', stop_words='english')

x_count_filter_train = count_filter_vec.fit_transform(x_train)
x_count_filter_test = count_filter_vec.transform(x_test)

x_tfidf_filter_train = tfidf_filter_vec.fit_transform(x_train)
x_tfidf_filter_test = tfidf_filter_vec.transform(x_test)

from sklearn.naive_bayes import MultinomialNB

mnb_count_filter = MultinomialNB()
mnb_count_filter.fit(x_count_filter_train, y_train)
y_count_filter_predict = mnb_count_filter.predict(x_count_filter_test)

mnb_tfidf_filter = MultinomialNB()
mnb_tfidf_filter.fit(x_tfidf_filter_train, y_train)
y_tfidf_filter_predict = mnb_tfidf_filter.predict(x_tfidf_filter_test)

from sklearn.metrics import classification_report
print(classification_report(y_test, y_count_filter_predict, target_names=news.target_names))
print(classification_report(y_test, y_tfidf_filter_predict, target_names=news.target_names))

运行结果如下：

precision recall f1-score support

alt.atheism 0.90 0.90 0.90 108
comp.graphics 0.62 0.88 0.73 130
comp.os.ms-windows.misc 0.95 0.22 0.36 163
comp.sys.ibm.pc.hardware 0.61 0.81 0.70 141
comp.sys.mac.hardware 0.87 0.86 0.87 145
comp.windows.x 0.72 0.91 0.81 141
misc.forsale 0.92 0.77 0.84 159
rec.autos 0.90 0.92 0.91 139
rec.motorcycles 0.94 0.95 0.94 153
rec.sport.baseball 0.96 0.91 0.93 141
rec.sport.hockey 0.94 0.97 0.95 148
sci.crypt 0.92 0.99 0.95 143
sci.electronics 0.88 0.83 0.86 160
sci.med 0.95 0.92 0.94 158
sci.space 0.89 0.94 0.92 149
soc.religion.christian 0.86 0.97 0.91 157
talk.politics.guns 0.85 0.96 0.90 134
talk.politics.mideast 0.95 0.99 0.97 133
talk.politics.misc 0.89 0.93 0.91 130
talk.religion.misc 0.98 0.61 0.75 97

avg / total 0.88 0.86 0.85 2829

precision recall f1-score support

alt.atheism 0.90 0.88 0.89 108
comp.graphics 0.80 0.86 0.83 130
comp.os.ms-windows.misc 0.91 0.76 0.83 163
comp.sys.ibm.pc.hardware 0.70 0.83 0.76 141
comp.sys.mac.hardware 0.92 0.88 0.90 145
comp.windows.x 0.86 0.88 0.87 141
misc.forsale 0.92 0.78 0.84 159
rec.autos 0.90 0.95 0.92 139
rec.motorcycles 0.92 0.95 0.94 153
rec.sport.baseball 0.95 0.94 0.94 141
rec.sport.hockey 0.91 0.99 0.95 148
sci.crypt 0.81 0.99 0.89 143
sci.electronics 0.92 0.80 0.86 160
sci.med 0.98 0.89 0.93 158
sci.space 0.88 0.95 0.91 149
soc.religion.christian 0.72 0.98 0.83 157
talk.politics.guns 0.85 0.94 0.89 134
talk.politics.mideast 0.94 1.00 0.97 133
talk.politics.misc 0.98 0.78 0.87 130
talk.religion.misc 1.00 0.35 0.52 97

avg / total 0.89 0.88 0.87 2829