tfidf 与 Python 的数据框

Question

我必须对一些情绪进行分类，我的数据框是这样的

Phrase                      Sentiment    
is it  good movie          positive    
wooow is it very goode      positive    
bad movie                  negative

我做了一些预处理，如标记化停止词词干等......我得到了

Phrase                      Sentiment    
[ good , movie  ]        positive    
[wooow ,is , it ,very, good  ]   positive 
[bad , movie ]            negative

我最终需要得到一个数据框，其中行是文本，值为 tf_idf，列是这样的词

good     movie   wooow    very      bad                Sentiment
tf idf    tfidf_  tfidf    tf_idf    tf_idf               positive
(same thing for the 2 remaining lines)

Answer 1

我会使用专为此类任务设计的sklearn.feature_extraction.text.TfidfVectorizer：

演示：

In [63]: df
Out[63]:
                   Phrase Sentiment
0       is it  good movie  positive
1  wooow is it very goode  positive
2               bad movie  negative

解决方案：

from sklearn.feature_extraction.text import CountVectorizer, TfidfVectorizer

vect = TfidfVectorizer(sublinear_tf=True, max_df=0.5, analyzer='word', stop_words='english')

X = vect.fit_transform(df.pop('Phrase')).toarray()

r = df[['Sentiment']].copy()

del df

df = pd.DataFrame(X, columns=vect.get_feature_names())

del X
del vect

r.join(df)

结果：

In [31]: r.join(df)
Out[31]:
  Sentiment  bad  good     goode     wooow
0  positive  0.0   1.0  0.000000  0.000000
1  positive  0.0   0.0  0.707107  0.707107
2  negative  1.0   0.0  0.000000  0.000000

更新：节省内存的解决方案：

from sklearn.feature_extraction.text import CountVectorizer, TfidfVectorizer

vect = TfidfVectorizer(sublinear_tf=True, max_df=0.5, analyzer='word', stop_words='english')

X = vect.fit_transform(df.pop('Phrase')).toarray()

for i, col in enumerate(vect.get_feature_names()):
    df[col] = X[:, i]

UPDATE2： related question where the memory issue was finally solved

Answer 2

设置

df = pd.DataFrame([
        [['good', 'movie'], 'positive'],
        [['wooow', 'is', 'it', 'very', 'good'], 'positive'],
        [['bad', 'movie'], 'negative']
    ], columns=['Phrase', 'Sentiment'])

df

                        Phrase Sentiment
0                [good, movie]  positive
1  [wooow, is, it, very, good]  positive
2                 [bad, movie]  negative

---

计算term frequency tf

# use `value_counts` to get counts of items in list
tf = df.Phrase.apply(pd.value_counts).fillna(0)
print(tf)

   bad  good   is   it  movie  very  wooow
0  0.0   1.0  0.0  0.0    1.0   0.0    0.0
1  0.0   1.0  1.0  1.0    0.0   1.0    1.0
2  1.0   0.0  0.0  0.0    1.0   0.0    0.0

计算inverse document frequency idf

# add one to numerator and denominator just incase a term isn't in any document
# maximum value is log(N) and minimum value is zero
idf = np.log((len(df) + 1 ) / (tf.gt(0).sum() + 1))
idf

bad      0.693147
good     0.287682
is       0.693147
it       0.693147
movie    0.287682
very     0.693147
wooow    0.693147
dtype: float64

---

tfidf

tdf * idf

        bad      good        is        it     movie      very     wooow
0  0.000000  0.287682  0.000000  0.000000  0.287682  0.000000  0.000000
1  0.000000  0.287682  0.693147  0.693147  0.000000  0.693147  0.693147
2  0.693147  0.000000  0.000000  0.000000  0.287682  0.000000  0.000000