回答編集履歴
1
追記
answer
CHANGED
|
@@ -5,4 +5,54 @@
|
|
|
5
5
|
n = 10 # いくつほしいか
|
|
6
6
|
feature_words = [feature_names[doc[:n]] for doc in index]
|
|
7
7
|
```
|
|
8
|
-
各文書ごとに特徴語を取り出したければ、こんな感じでいけると思います。
|
|
8
|
+
各文書ごとに特徴語を取り出したければ、こんな感じでいけると思います。
|
|
9
|
+
|
|
10
|
+
### 追記
|
|
11
|
+
せっかくなので簡単なサンプルを。
|
|
12
|
+
|
|
13
|
+
```python
|
|
14
|
+
import numpy as np
|
|
15
|
+
from sklearn.feature_extraction.text import TfidfVectorizer
|
|
16
|
+
from sklearn.datasets import fetch_20newsgroups
|
|
17
|
+
|
|
18
|
+
news20 = fetch_20newsgroups()
|
|
19
|
+
vectorizer = TfidfVectorizer(min_df=0.03)
|
|
20
|
+
|
|
21
|
+
tfidf = vectorizer.fit_transform(news20.data[:1000]).toarray()
|
|
22
|
+
feature_names = np.array(vectorizer.get_feature_names())
|
|
23
|
+
index = tfidf.argsort(axis=1)[:,::-1]
|
|
24
|
+
feature_words = [feature_names[doc] for doc in index]
|
|
25
|
+
|
|
26
|
+
n = 5 # top何単語取るか
|
|
27
|
+
m = 10 # 何記事サンプルとして抽出するか
|
|
28
|
+
targets = np.array(news20.target_names)[news20.target[:m]]
|
|
29
|
+
|
|
30
|
+
for fwords, target in zip(feature_words, targets):
|
|
31
|
+
print(target)
|
|
32
|
+
print(fwords[:n])
|
|
33
|
+
|
|
34
|
+
""" =>
|
|
35
|
+
rec.autos
|
|
36
|
+
['car' 'was' 'this' 'the' 'where']
|
|
37
|
+
comp.sys.mac.hardware
|
|
38
|
+
['washington' 'add' 'guy' 'speed' 'call']
|
|
39
|
+
comp.sys.mac.hardware
|
|
40
|
+
['the' 'display' 'anybody' 'heard' 'disk']
|
|
41
|
+
comp.graphics
|
|
42
|
+
['division' 'chip' 'systems' 'computer' 'four']
|
|
43
|
+
sci.space
|
|
44
|
+
['error' 'known' 'tom' 'memory' 'the']
|
|
45
|
+
talk.politics.guns
|
|
46
|
+
['of' 'the' 'com' 'to' 'says']
|
|
47
|
+
sci.med
|
|
48
|
+
['thanks' 'couldn' 'instead' 'file' 'everyone']
|
|
49
|
+
comp.sys.ibm.pc.hardware
|
|
50
|
+
['chip' 'is' 'fast' 'ibm' 'bit']
|
|
51
|
+
comp.os.ms-windows.misc
|
|
52
|
+
['win' 'help' 'please' 'appreciated' 'figure']
|
|
53
|
+
comp.sys.mac.hardware
|
|
54
|
+
['the' 'file' 'lost' 've' 'it']
|
|
55
|
+
"""
|
|
56
|
+
```
|
|
57
|
+
|
|
58
|
+
それっぽく動いているようです。
|