unsupervised_topic_segmentation icon indicating copy to clipboard operation
unsupervised_topic_segmentation copied to clipboard
verified publisher icon gdamaskinos

potential bug?

Open chijames opened this issue 4 years ago • 2 comments

Hi,

Thanks for releasing the code.

Just wonder if we need to add textiling_hyperparameters.SENTENCE_COMPARISON_WINDOW to the predicted segments? The first possible segmentation should occur at textiling_hyperparameters.SENTENCE_COMPARISON_WINDOW, right?

Thanks.

chijames avatar Jul 28 '21 01:07 chijames

yes, I also think the same. Looks like the code is not complete. If first predicted segment is at 0 then the topic break should be at (0 + extiling_hyperparameters.SENTENCE_COMPARISON_WINDOW)th sentence in the original document

Akshayextreme avatar Dec 23 '21 12:12 Akshayextreme

@Akshayextreme I think it should be at 0 + extiling_hyperparameters.SENTENCE_COMPARISON_WINDOW + extiling.SMOOTHING_WINDOW, I am currently in the debugger and while the depth score computation removes SENTENCE_COMPARISON_WINDOW at either end, it removes extiling.SMOOTHING_WINDOW at either end as well

aanchan avatar Oct 11 '22 18:10 aanchan