• Àüü
  • ÀüÀÚ/Àü±â
  • Åë½Å
  • ÄÄÇ»ÅÍ
´Ý±â

»çÀÌÆ®¸Ê

Loading..

Please wait....

±¹³» ³í¹®Áö

Ȩ Ȩ > ¿¬±¸¹®Çå > ±¹³» ³í¹®Áö > Çѱ¹Á¤º¸°úÇÐȸ ³í¹®Áö > Á¤º¸°úÇÐȸ ÄÄÇ»ÆÃÀÇ ½ÇÁ¦ ³í¹®Áö (KIISE Transactions on Computing Practices)

Á¤º¸°úÇÐȸ ÄÄÇ»ÆÃÀÇ ½ÇÁ¦ ³í¹®Áö (KIISE Transactions on Computing Practices)

Current Result Document :

ÇѱÛÁ¦¸ñ(Korean Title) »çÀüÇнÀ ¾ð¾î¸ðµ¨ ±â¹ÝÀÇ Çѱ¹¾î Áú¹®-´äº¯ µ¥ÀÌÅÍ Áõ°­ ¹æ¹ý
¿µ¹®Á¦¸ñ(English Title) A Pretrained Language Model-Based Data Augmentation Method for Korean Question-Answering Systems
ÀúÀÚ(Author) Á¶¿ìÁø   ÀÌÇõÁØ   Woojin Cho   Hyukjoon Lee  
¿ø¹®¼ö·Ïó(Citation) VOL 27 NO. 12 PP. 0563 ~ 0573 (2021. 11)
Çѱ۳»¿ë
(Korean Abstract)
ÀÚ¿¬¾î󸮴 ÃÖ±Ù ÀΰøÁö´ÉÀÌ °¢±¤À» ¹ÞÀ¸¸ç ºñ¾àÀûÀÎ ¹ßÀüÀ» ÀÌ·ç°í ÀÖ´Ù. ÀÚ¿¬¾îó¸®ÀÇ ¿©·¯ ¹®Á¦ Áß Áú¹®-´äº¯Àº ÀΰøÁö´ÉÀÌ ¹®´Ü ³»¿¡¼­ Áú¹®¿¡ ¸Â´Â ´äÀ» ã¾ÆÁÖ´Â ¹®Á¦´Ù. ÀΰøÁö´É ¹®Á¦¿¡¼­ ¿ì¼öÇÑ ¼º´ÉÀ» ´Þ¼ºÇϱâ À§Çؼ­´Â ÀΰøÁö´É ¸ðµ¨°ú ÇнÀ µ¥ÀÌÅͼÂÀÇ È®º¸°¡ ¸Å¿ì Áß¿äÇϴ٠ƯÈ÷ Áú¹®-´äº¯ µ¥ÀÌÅͼÂÀº Áú¹®-´äº¯ÀÇ ¹®¹ý, °ü°è µî Àΰ£ÀÇ Á÷Á¢Àû °³ÀÔÀÌ ¸¹ÀÌ ¿ä±¸µÇ¾î µ¥ÀÌÅÍ ±¸ÃàÀÌ ½±Áö ¾Ê´Ù. ÀÌ·± ¹®Á¦Á¡À» ÇØ°áÇϱâ À§ÇØ º» ³í¹®¿¡¼­´Â ´äº¯ »ý¼º, Áú¹® »ý¼º, ÇÊÅ͸µÀÇ 3 ´Ü°è·Î ±¸¼ºµÈ Áú¹®-´äº¯ µ¥ÀÌÅÍ Áõ°­ ¹æ¹ýÀ» Á¦¾ÈÇÑ´Ù. Áõ°­µÈ µ¥ÀÌÅ͸¦ ÀÌ¿ëÇÏ¿© ÇнÀ½ÃŲ ¸ðµ¨ÀÇ ÁúÀÇÀÀ´ä ¼º´ÉÀÌ KorQuAD µ¥ÀÌÅ͸¸À¸·Î ÇнÀ½ÃŲ ¸ðµ¨¿¡ ºñÇØ F1-score ±âÁØ ÃÖ´ë 1.13 Áõ°¡ÇÑ °á°ú¸¦ ¾òÀ» ¼ö ÀÖÀ½À» ½ÇÇèÀ» ÅëÇØ º¸ÀδÙ.
¿µ¹®³»¿ë
(English Abstract)
Abstract Natural language processing (NLP) has recently made rapid progress with artificial intelligence (AI) in the spotlight. Among the many problems of NLP, question-answer (QA) is a problem in which an AI algorithm finds the right answer to the question within a paragraph. Securing artificial intelligence models and training data are utmost important to achieve good performance of AI. In particular, QA data requires a lot of direct human intervention due to grammars and relationships between questions and answers, making it difficult to obtain a data set. To solve this problem, this paper proposes a QA data augmentation method consisting of four steps: answer generation, question generation, round-trip filter technique, and verification. Experiment results shows that the QA performance of the model trained using the augmented data could achieve up to 1.13-fold increase in terms of F1-score compared to the model learned by using KorQuAD data only.
Å°¿öµå(Keyword) ÀÚ¿¬¾î󸮠  Áú¹®-´äº¯   KorQuAD   ¸»¹¶Ä¡   ÀÚ¿¬¾î»ý¼º   natural language processing   question-answering   KorQuAD   Korean corpus   natural language generation  
ÆÄÀÏ÷ºÎ PDF ´Ù¿î·Îµå