• 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) ¸®ºä ±â¹ÝÀÇ ¾îÅÙ¼Ç ¸ÞÄ¿´ÏÁò °³¼±
¿µ¹®Á¦¸ñ(English Title) Improving Review-based Attention Mechanism
ÀúÀÚ(Author) À¯¼º¿í   ±¸ÇÑÁØ   ½É±Ô¼®   Sungwook Yoo   Hanjun Goo   Kyuseok Shim  
¿ø¹®¼ö·Ïó(Citation) VOL 27 NO. 10 PP. 0486 ~ 0491 (2021. 10)
Çѱ۳»¿ë
(Korean Abstract)
Ãßõ½Ã½ºÅÛÀº À¯ÀúÀÇ ¼Òºñ¸¦ À¯µµÇϱâ À§ÇØ ´Ù¾çÇÑ Ç÷§Æû¿¡¼­ »ç¿ëµÇ°í ÀÖ´Ù. ÃÖ±Ù ÆòÁ¡À» Àß ¿¹ÃøÇϱâ À§ÇÏ¿© À¯Àú°¡ ¾ÆÀÌÅÛ¿¡ ´ëÇÏ¿© ÀÛ¼ºÇÑ ¸®ºä¸¦ È°¿ëÇÏ´Â ¿¬±¸µéÀÌ ÀÖ¾ú´Ù. ÀÌ Áß ÅؽºÆ®¿¡ ¾îÅÙ¼Ç ¸ÞÄ¿´ÏÁòÀ» Àû¿ëÇÏ¿© À¯ÀúÀÇ ¼±È£µµ¿Í ¾ÆÀÌÅÛÀÇ Æ¯Â¡À» ÆľÇÇÏ´Â °ÍÀÌ È¿°úÀûÀ̾ú´Ù. ÇÏÁö¸¸, ´Ü¾î ¼öÁØÀÇ ¾îÅÙ¼Ç ¸ÞÄ¿´ÏÁòÀº À¯ÀúÀÇ ¼±È£µµ¿Í ¾ÆÀÌÅÛÀÇ Æ¯¼ºÀ» Àß µå·¯³»´Â ¼­¼úÀûÀÎ ¸®ºä¸¦ ºÐ°£ÇÏÁö ¸øÇÑ´Ù. À¯Àú°¡ ¾ÆÀÌÅÛ¿¡ ´ëÇØ ÀÚ¼¼ÇÑ ¼³¸íÀ» ÇÏ´Â ¼­¼úÀûÀÎ ¸®ºä´Â »ó´ëÀûÀ¸·Î ´Ù¸¥ ŸÀÔÀÇ Çǵå¹é Á¤º¸º¸´Ù Áß¿äÇÏ´Ù. µû¶ó¼­, º» ³í¹®Àº ´Ü¾î ¼öÁØÀ¸·Î °³º°Àû ¿äÀÎÀ» °í·ÁÇÏ¿© Áß¿äµµ¸¦ ¹Ý¿µÇÏ´Â ¹æ½Ä°ú ¸®ºä ¼öÁØÀ¸·Î ¼­¼úÀû Áß¿äµµ¸¦ ¹Ý¿µÇÏ´Â ¹æ½ÄÀ» °èÃþÀûÀ¸·Î À¶ÇÕÇÏ¿© »ç¿ëÀÚ ¼±È£µµ º¤ÅÍ¿Í ¾ÆÀÌÅÛ Æ¯Â¡ º¤Å͸¦ ÃßÃâÇÏ´Â ¹æ½ÄÀ» Á¦¾ÈÇÑ´Ù. ÇÏÁö¸¸ µ¥ÀÌÅÍ Èñ¼Ò¼º Çö»ó ¶§¹®¿¡ ¸ðµ¨ ÇнÀÀÌ ÈÆ·Ã µ¥ÀÌÅÍ¿¡ °úÀûÇÕ µÉ ¼ö ÀÖ´Ù. ÀÌ ¹®Á¦¸¦ º¸¿ÏÇϱâ À§ÇØ È®·üÀû °¡ÁßÄ¡ Æò±Õ ¹æ¹ýÀ» Àû¿ëÇÏ¿© ¸ðµ¨ÀÇ ÀϹÝÈ­Çϴµ¥ µµ¿òÀ» ÁÖ¾ú´Ù. 3°¡Áö ½Ç»ýÈ° µ¥ÀÌÅÍ¿¡ ´ëÇØ º» ³í¹®¿¡¼­ Á¦¾ÈÇÏ´Â ¸ðµ¨ÀÇ ¼º´ÉÀÌ ¿ì¼öÇÔÀ» °ËÁõÇÏ¿´´Ù.
¿µ¹®³»¿ë
(English Abstract)
The recommender system is used on various platforms to improve user consumption. Recent studies have attempted to incorporate review information into rating prediction in order to increase accuracy of rating prediction. Among them, it was efficient to apply the attention mechanism to text information by focusing important words to determine preference of a user and property of an item. However, the word-level attention mechanism did not discern descriptive reviews that reveal user¡¯s preference and item property. Descriptive reviews, in which users provided extensive descriptions of items, were more important than other types of feedback information. As a result, our model hierarchically integrated the word-level attention mechanism, which took individual factors into consideration, and the review-level attention mechanism, which reflected descriptive utility of reviews. However, due to the data sparsity problem, the trained model may be biased. To overcome this issue, we applied stochastic weight average technique for boosting generalization of our model. We showed the effectiveness of our proposed model by conducting experiments with three real-life datasets.
Å°¿öµå(Keyword) Ãßõ½Ã½ºÅÛ   ÄÁº¼·ç¼Ç ½Å°æ¸Á   ¾îÅÙ¼Ç ¸ÞÄ¿´ÏÁò   recommender system   convolutional neural network   attention mechanism  
ÆÄÀÏ÷ºÎ PDF ´Ù¿î·Îµå