Á¤º¸°úÇÐȸ³í¹®Áö (Journal of KIISE)
ÇѱÛÁ¦¸ñ(Korean Title) |
´Ü¾î »ý¼º ÀÌ·ÂÀ» ÀÌ¿ëÇÑ ¿ä¾à¹® »ý¼ºÀÇ ¾îÈÖ ¹Ýº¹ ¹®Á¦ ÇØ°á |
¿µ¹®Á¦¸ñ(English Title) |
Solving for Redundant Repetition Problem of Generating Summarization using Decoding History |
ÀúÀÚ(Author) |
·ùÀçÇö
³ëÀ±¼®
ÃÖ¼öÁ¤
¹Ú¼¼¿µ
¹Ú¼º¹è
Jaehyun Ryu
Yunseok Noh
Su Jeong Choi
Seyoung Park
Seong-Bae Park
|
¿ø¹®¼ö·Ïó(Citation) |
VOL 46 NO. 06 PP. 0535 ~ 0543 (2019. 06) |
Çѱ۳»¿ë (Korean Abstract) |
½ÃÄö½º-Åõ-½ÃÄö½º ±â¹ÝÀÇ ¿ä¾à ¸ðµ¨¿¡¼ ÀÚÁÖ ¹ß»ýÇÏ´Â ¹®Á¦ Áß Çϳª´Â ¿ä¾à¹®ÀÇ »ý¼º°úÁ¤¿¡¼ ´Ü¾î³ª ±¸, ¹®ÀåÀÌ ºÒÇÊ¿äÇÏ°Ô ¹Ýº¹ÀûÀ¸·Î »ý¼ºµÇ´Â °ÍÀÌ´Ù. À̸¦ ÇØ°áÇϱâ À§ÇØ ±âÁ¸ ¿¬±¸µéÀº ´ëºÎºÐ ¸ðµ¨¿¡ ¿©·¯ ¸ðµâÀ» Ãß°¡ÇÏ´Â ¹æ¹ýÀ» Á¦¾ÈÇßÁö¸¸, À§ ¹æ¹ýÀº »ý¼ºÇÏÁö ¸»¾Æ¾ß ÇÏ´Â ´Ü¾î¿¡ ´ëÇÑ ÇнÀÀÌ ºÎÁ·ÇÏ¿© ¹Ýº¹ »ý¼º ¹®Á¦¸¦ ÇØ°áÇÔ¿¡ ÀÖ¾î ÇÑ°è°¡ ÀÖ´Ù. º» ³í¹®¿¡¼´Â ´Ü¾î »ý¼º ÀÌ·ÂÀ» Á÷Á¢ÀûÀ¸·Î ÀÌ¿ëÇÏ¿© ¹Ýº¹ »ý¼ºÀ» Á¦¾îÇÏ´Â Repeat Loss¸¦ ÀÌ¿ëÇÑ »õ·Î¿î ÇнÀ ¹æ¹ýÀ» Á¦¾ÈÇÑ´Ù. Repeat Loss¸¦ µðÄÚ´õ°¡ ´Ü¾î »ý¼º È®·üÀ» °è»ê ÇßÀ» ¶§ ÀÌÀü¿¡ »ý¼ºÇÑ ´Ü¾î°¡ ´Ù½Ã »ý¼ºµÉ È®·ü·Î Á¤ÀÇÇÔÀ¸·Î½á ½ÇÁ¦ »ý¼ºÇÑ ´Ü¾î°¡ ¹Ýº¹ »ý¼ºµÉ È®·üÀ» Á÷Á¢ÀûÀ¸·Î Á¦¾îÇÒ ¼ö ÀÖ´Ù. Á¦¾ÈÇÑ ¹æ¹ýÀ¸·Î ¿ä¾à ¸ðµ¨À» ÇнÀÇÑ °á°ú, ´Ü¾î ¹Ýº¹ÀÌ ÁÙ¾îµé¾î ¾çÁúÀÇ ¿ä¾àÀ» »ý¼ºÇÏ´Â °ÍÀ» ½ÇÇèÀûÀ¸·Î È®ÀÎÇÒ ¼ö ÀÖ¾ú´Ù.
|
¿µ¹®³»¿ë (English Abstract) |
Neural attentional sequence-to-sequence models have achieved great success in abstractive summarization. However, the model is limited by several challenges including repetitive generation of words, phrase and sentences in the decoding step. Many studies have attempted to address the problem by modifying the model structure. Although the consideration of actual history of word generation is crucial to reduce word repetition, these methods, however, do not consider the decoding history of generated sequence. In this paper, we propose a new loss function, called ¡®Repeat Loss¡¯ to avoid repetitions. The Repeat Loss directly prevents the model from repetitive generation of words by giving a loss penalty to the generation probability of words already generated in the decoding history. Since the propose Repeat Loss does not need a special network structure, the loss function is applicable to any existing sequence-to-sequence models. In experiments, we applied the Repeat Loss to a number of sequence-to-sequence model based summarization systems and trained them on both Korean and CNN/Daily Mail summarization datasets. The results demonstrate that the proposed method reduced repetitions and produced high-quality summarization.
|
Å°¿öµå(Keyword) |
¹®¼ ¿ä¾à
¹Ýº¹ Á¦¾î
½ÃÄö½º-Åõ-½ÃÄö½º
¼Õ½Ç ÇÔ¼ö
text summarization
sequence-to-sequence model
word repetition
repeat loss
|
ÆÄÀÏ÷ºÎ |
PDF ´Ù¿î·Îµå
|