def textRank(tokenized_words, tag_word_dict):
words_set = syntactic_filter(tag_word_dict)
# Add the vertex to the graph
graph_dict = {}
for w in words_set:
graph_dict[w] = []
graph = Graph(graph_dict)
# Add the edges
words = nltk.Text (tokenized_words)
doc = nltk.ConcordanceIndex(words)
for w in words_set:
results = get_concordance(w, doc)
for context in results:
left = context[0].split()
right = context[1].split()
for l in left:
if l in words_set:
graph.add_edge((w, l))
graph.add_edge((l, w))
for l in right:
if l in words_set:
graph.add_edge((w, l))
graph.add_edge((l, w))
# Run the text rank algorithm
delta = 1
i = 0
d = 0.85
while (delta > 0.0001 and i < 5000):
for v in graph.vertices():
degree = graph.vertex_degree(v)
old_rank = graph.text_rank(v)
sum = 0
for v2 in graph.adjacency_list(v):
degree2 = graph.vertex_degree(v2)
# print ("Degree for " + v2 + " = " + str(degree2))
tr = graph.text_rank(v2)
sum += tr / degree2
value = (1 - d) + d * sum
graph.set_text_rank(v, value)
if abs(value - old_rank) < delta:
delta = abs(value - old_rank)
i = i + 1
text_rank_dict = {}
for v in graph.vertices():
text_rank_dict[v] = graph.text_rank(v)
sorted_text_rank = sorted(text_rank_dict.items(), key=operator.itemgetter(1), reverse=True)
return sorted_text_rank
