A question about computing term frequency in Lexrank

[IsakZhang]

Hi, i read the code of computing term frequency in LexRank metrics[term] = tf / max_tf.
I don't understand why do u divide the maximum count of word in that sentence rather than use the number of occurrences of the word in the sentence(described in the original lexrank paper), or just normalize it by dividing total count of all words?

[miso-belica]

@IsakZhang Hi, it's tough question for me. I really don't remember why and if I diverted from the original paper. But I usually did such thing because I inspired somewhere else and it gave me the better results or I just did some experiments with better results.