Get the top term per document - scikit tf-idf

Question

Afte vectorizing multiple documents with scikit's tf-idf vectorizer , is there a way to get the most 'influential' term per document?

I have only found ways of getting the most 'influential' terms for the entire corpus, not for each document, though.

Answer 1

Just adding one more way of doing this, in the last two steps of Ami :

# Get a list of all the keywords by calling function
feature_names = np.array(count_vect.get_feature_names())
feature_names[X_train_tfidf.argmax(axis=1)]

Answer 2

Say you start with a dataset:

from sklearn.feature_extraction.text import CountVectorizer
from sklearn.feature_extraction.text import TfidfTransformer
import numpy as np
from sklearn.datasets import fetch_20newsgroups

d = fetch_20newsgroups()

Use a count vectorizer and tfidf:

count_vect = CountVectorizer()
X_train_counts = count_vect.fit_transform(d.data)
transformer = TfidfTransformer()
X_train_tfidf = transformer.fit_transform(X_train_counts)

Now you can create an inverse mapping:

m = {v: k for (k, v) in count_vect.vocabulary_.items()}

and this gives the influential word per doc:

[m[t] for t in np.array(np.argmax(X_train_tfidf, axis=1)).flatten()]