# -*- coding: utf-8 -*-
from __future__ import division
import networkx
from .baseclass import BaseSummarizer
class TextRankSummarizer(BaseSummarizer):
def summarize(self, text, length=5, weighting='frequency', norm=None):
"""
Implements the TextRank summarization algorithm, which follows closely to the PageRank algorithm for ranking
web pages.
:param text: a string of text to be summarized, path to a text file, or URL starting with http
:param length: the length of the output summary; either a number of sentences (e.g. 5) or a percentage
of the original document (e.g. 0.5)
:param weighting: 'frequency', 'binary' or 'tfidf' weighting of sentence terms ('frequency' by default)
:param norm: if 'l1' or 'l2', normalizes words by the length of their associated sentence to "down-weight"
the voting power of long sentences (None by default)
:return: list of sentences for the summary
"""
text = self._parse_input(text)
sentences, unprocessed_sentences = self._tokenizer.tokenize_sentences(text)
length = self._parse_summary_length(length, len(sentences))
if length == len(sentences):
return unprocessed_sentences
# Compute the word frequency matrix. If norm is set to 'l1' or 'l2' then words are normalized
# by the length of their associated sentences (such that each vector of sentence terms sums to 1).
word_matrix = self._compute_matrix(sentences, weighting=weighting, norm=norm)
# Build the similarity graph by calculating the number of overlapping words between all
# combinations of sentences.
similarity_matrix = (word_matrix * word_matrix.T)
similarity_graph = networkx.from_scipy_sparse_matrix(similarity_matrix)
scores = networkx.pagerank(similarity_graph)
ranked_sentences = sorted(
((score, ndx) for ndx, score in scores.items()), reverse=True
)
top_sentences = [ranked_sentences[i][1] for i in range(length)]
top_sentences.sort()
return [unprocessed_sentences[i] for i in top_sentences]
