Á¤º¸Ã³¸®ÇÐȸ ³í¹®Áö ÄÄÇ»ÅÍ ¹× Åë½Å½Ã½ºÅÛ
ÇѱÛÁ¦¸ñ(Korean Title) |
È¿°úÀûÀÎ ¹®¼ ¼öÁØÀÇ Á¤º¸¸¦ ÀÌ¿ëÇÑ ÇÕ¼º°ö ½Å°æ¸Á ±â¹ÝÀÇ ½Å±Ô¼º ŽÁö |
¿µ¹®Á¦¸ñ(English Title) |
CNN-Based Novelty Detection with Effectively Incorporating Document-Level Information |
ÀúÀÚ(Author) |
Á¶¼º¿õ
¿ÀÈï¼±
ÀÓ»óÈÆ
±è¼±È£
Seongung Jo
Heung-Seon Oh
Sanghun Im
Seonho Kim
|
¿ø¹®¼ö·Ïó(Citation) |
VOL 09 NO. 10 PP. 0231 ~ 0238 (2020. 10) |
Çѱ۳»¿ë (Korean Abstract) |
À¥ »ó¿¡ ¼ö ¸¹Àº ¹®¼°¡ µîÀåÇÔ¿¡ µû¶ó ±âÁ¸ ¹®¼¿Í ³»¿ëÀÌ Áߺ¹µÇ´Â ¹®¼¸¦ ã¾Æ¼ Á¦¿ÜÇÔÀ¸·Î½á »õ·Î¿î ¹®¼¸¦ ã´Â ³ë·ÂÀ» ÁÙÀÏ ¼ö ÀÖ¾î ¹®¼ ¼öÁØÀÇ ½Å±Ô¼º ŽÁö(novelty detection)°¡ Áß¿äÇØÁ³´Ù. ÃÖ±Ù ¿¬±¸¿¡¼´Â ÇÕ¼º°ö ½Å°æ¸Á(CNN) ±¸Á¶ ±â¹ÝÀÇ ½Å±Ô¼º ŽÁö ¸ðµ¨ ±¸Á¶°¡ Á¦¾ÈµÇ¾ú°í »ó´çÇÑ ¼º´É Çâ»óÀ» ³ªÅ¸³»¿´´Ù. º» ³í¹®¿¡¼´Â ±âÁ¸ÀÇ CNN ±â¹ÝÀÇ ¸ðµ¨¿¡¼ ¹®¼ ¼öÁØÀÇ Á¤º¸°¡ Á¦ÇÑÀûÀ¸·Î »ç¿ëµÇ´Â °ÍÀ» °üÃøÇÏ°í ¹®¼ÀÇ ½Å±Ô¼ºÀ» °áÁ¤ÇÒ ¶§ ¹®¼ ¼öÁØÀÇ Á¤º¸°¡ Áß¿äÇϹǷΠÁ¦ÇÑÀûÀÎ »ç¿ëÀÌ ¹®Á¦°¡ µÈ´Ù°í °¡Á¤ÇÏ¿´´Ù. ÀÌ¿¡ ´ëÇÑ ÇØ°áÃ¥À¸·Î, º» ³í¹®¿¡¼´Â ÇÕ¼º°ö ½Å°æ¸Á ±â¹Ý ½Å±Ô¼º ŽÁö ¸ðµ¨ ±¸Á¶¸¦ °³¼±ÇÏ¿© ¹®¼ ¼öÁØ Á¤º¸¸¦ È¿°úÀûÀ¸·Î »ç¿ëÇÏ´Â µÎ °¡Áö ¹æ¹ýÀ» Á¦¾ÈÇÑ´Ù. º» ³í¹®¿¡¼ Á¦¾ÈÇÏ´Â ¹æ¹ýÀº ´ë»ó(target) ¹®¼¿Í Áõ°Å·Î ÁÖ¾îÁø Ãâó(source) ¹®¼ »çÀÌÀÇ »ó´ëÀû(relative) Á¤º¸¸¦ ÃßÃâÇÏ¿© ½Å±Ô¼ºÀ» ºÐ·ùÇÒ ´ë»ó ¹®¼ÀÇ Æ¯Â¡ º¤Å͸¦ ±¸¼ºÇÏ´Â °Í¿¡ ÃÊÁ¡À» ¸ÂÃá´Ù. º» ³í¹®¿¡¼´Â Ç¥ÁØ º¥Ä¡¸¶Å© µ¥ÀÌÅÍ ¼ÂÀÎ TAP-DLND 1.0¸¦ ÀÌ¿ëÇÏ¿© ¿©·¯ ½ÇÇèÀ» ÅëÇؼ Á¦¾ÈÇÑ ¹æ¹ýÀÇ ¿ì¼ö¼ºÀ» º¸¿©ÁØ´Ù
|
¿µ¹®³»¿ë (English Abstract) |
With a large number of documents appearing on the web, document-level novelty detection has become important since it can reduce the efforts of finding novel documents by discarding documents sharing redundant information already seen. A recent work proposed a convolutional neural network (CNN)-based novelty detection model with significant performance improvements. We observed that it has a restriction of using document-level information in determining novelty but assumed that the document-level information is more important. As a solution, this paper proposed two methods of effectively incorporating document-level information using a CNN-based novelty detection model. Our methods focus on constructing a feature vector of a target document to be classified by extracting relative information between the target document and source documents given as evidence. A series of experiments showed the superiority of our methods on a standard benchmark collection, TAP-DLND 1.0.
|
Å°¿öµå(Keyword) |
µö ·¯´×
ÇÕ¼º°ö ½Å°æ¸Á
½Å±Ô¼º ŽÁö
Deep Learning
CNN
Novelty Detection
|
ÆÄÀÏ÷ºÎ |
PDF ´Ù¿î·Îµå
|