Get the document name in scikit-learn tf-idf matrix

问题

I have created a tf-idf matrix but now I want to retrieve top 2 words for each document. I want to pass document id and it should give me the top 2 words.

Right now, I have this sample data:

from sklearn.feature_extraction.text import TfidfVectorizer

d = {'doc1':"this is the first document",'doc2':"it is a sunny day"} ### corpus

test_v = TfidfVectorizer(min_df=1)    ### applied the model
t = test_v.fit_transform(d.values())
feature_names = test_v.get_feature_names() ### list of words/terms

>>> feature_names
['day', 'document', 'first', 'is', 'it', 'sunny', 'the', 'this']

>>> t.toarray()
array([[ 0.        ,  0.47107781,  0.47107781,  0.33517574,  0.        ,
     0.        ,  0.47107781,  0.47107781],
   [ 0.53404633,  0.        ,  0.        ,  0.37997836,  0.53404633,
     0.53404633,  0.        ,  0.        ]])

I can access the matrix by giving the row number eg.

 >>> t[0,1]
   0.47107781233161794

Is there a way I can be able to access this matrix by document id? In my case 'doc1' and 'doc2'.

Thanks

回答1:

By doing

t = test_v.fit_transform(d.values())

you lose any link to the document ids. A dict is not ordered so you have no idea which value is given in which order. The means that before passing the values to the fit_transform function you need to record which value corresponds to which id.

For example what you can do is:

counter = 0
values = []
key = {}


for k,v in d.items():
    values.append(v)
    key[k] = counter
    counter+=1

t = test_v.fit_transform(values)

From there you can build a function to access this matix by document id:

def get_doc_row(docid):
    rowid = key[docid]
    row = t[rowid,:]
    return row

来源：https://stackoverflow.com/questions/26304191/get-the-document-name-in-scikit-learn-tf-idf-matrix