Axis-based alignment of scholarly papers and its presentation slides considering document structure

Abstract

Recently, most researchers make a presentation with presentation slides to introduce a paper in academic conferences. We can often retrieve and browse papers and presentation slides through websites. We consider that we can obtain information efficiently by using both of them, and we propose a method to align papers and its presentation slides at the fine granularity. Though there are some existing works on this alignment, our system tried to achieve better accuracy for this problem by proposing the two approaches: 1) the adjustment by axis alignments; and 2) two-step alignment. The content similarity between each slide and paragraph is unstable due to the small amount of texts in the slides. Therefore we also calculate the content similarity in section-level and consider the ancillary alignment in section-level. Also, we succeed to obtain better alignments by adjusting the alignment score or narrowing down the alignment candidates. Finally, for each slide, we calculate the scores of each paragraph, and determine the alignment according to the scores by associating the slide to a sequence of paragraphs. We created a small dataset manually and conducted an experiment to confirm the effectiveness of the proposed method.