Á¤º¸°úÇÐȸ ³í¹®Áö B : ¼ÒÇÁÆ®¿þ¾î ¹× ÀÀ¿ë
ÇѱÛÁ¦¸ñ(Korean Title) |
º£ÀÌÁö¾È Ã߷иÁÀ» ÀÌ¿ëÇÑ °Ë»ö¿£Áø ¼¼ºÎ ¸ðµâÀÇ »ó¼¼ ºÐ¼® ¹æ¹ý |
¿µ¹®Á¦¸ñ(English Title) |
Methodology for Analyzing Search Engine Modules using Bayesian Inference Network |
ÀúÀÚ(Author) |
¼Û»ç±¤
À̽¿ì
Á¤ÇѹÎ
Sa-kwang Song
Seungwoo Lee
Hanmin Jung
|
¿ø¹®¼ö·Ïó(Citation) |
VOL 40 NO. 05 PP. 0277 ~ 0282 (2013. 05) |
Çѱ۳»¿ë (Korean Abstract) |
°Ë»ö¿£ÁøÀÇ ¼º´ÉÀ» ³ôÀ̱â À§ÇØ °è¼ÓÀûÀ¸·Î ¼¼ºÎ ¸ðµâÀÇ ¼öÇà°á°ú¸¦ ºÐ¼®ÇÏ°í ¿À·ù¸¦ ¼öÁ¤ÇÏ´Â °úÁ¤À» °ÅÄ¡°Ô µÇ´Âµ¥, ÀÌ´Â ¸Å¿ì ½Ã°£°ú ³ë·ÂÀÌ ¸¹ÀÌ µé¾î°¡´Â ÀÛ¾÷ÀÌ´Ù. ±×·¯³ª, ÀÌ·¯ÇÑ °úÁ¤À» °ÅÃÄ ´ë»ó ¹®ÇåÀ¸·ÎºÎÅÍ »öÀξ ÃßÃâÇÏ´Â ¹æ¹ýÀ» °³¼±ÇÏ´Â °ÍÀº °Ë»ö¿£Áø ¼º´ÉÀ» ³ôÀ̱â À§ÇÑ ¸Å¿ì Áß¿äÇÑ ¿¬±¸ÁÖÁ¦ÀÌ´Ù. ÀϹÝÀûÀ¸·Î, ºÒ¿ë¾î ¸®½ºÆ®¸¦ ÅëÇØ ¼º´É¿¡ ±àÁ¤ÀûÀÎ ¿µÇâÀ» ¹ÌÄ¡Áö ¾Ê´Â »öÀξ Á¦°ÅÇϰųª, ÇÙ½É¾î ¶Ç´Â Àü¹®¿ë¾î µî »ó´ëÀûÀ¸·Î Áß¿äÇÑ »öÀξ °Á¶ÇÏ´Â ¹æ½ÄÀ» »ç¿ëÇÏ¿© °Ë»ö¿£ÁøÀÇ ¼º´ÉÀ» Çâ»ó½ÃŲ´Ù. ÇÏÁö¸¸, ¾îÀý ºÐ¸®, ÇüÅÂ¼Ò ºÐ¼®, ºÒ¿ë¾î ó¸® µî °Ë»ö¿£Áø ³»ºÎÀÇ ´Ù¾çÇÑ ´Ü°èº° ó¸® °úÁ¤À» °³¹ßÀÚ ¶Ç´Â ¿¬±¸ÀÚ°¡ °³º°Àû/Á¾ÇÕÀûÀ¸·Î »ìÆ캸°í ¼º´ÉÀ» °³¼±Çϱâ¶õ ³Ê¹«³ª ¾î·Á¿î ÀÛ¾÷ÀÌ´Ù. µû¶ó¼, º» ¿¬±¸¿¡¼´Â ¿¬±¸ÀÚ·Î ÇÏ¿©±Ý °³º°ÀûÀÎ »öÀξ °Ë»ö¿£Áø¿¡ ¹ÌÄ¡´Â ¿µÇâÀ» °¢ ´Ü°èº°·Î ¼Õ½±°Ô ºÐ¼®ÇÒ ¼ö ÀÖµµ·Ï, º£ÀÌÁö¾È Ã߷иÁ°ú ¾îÀýÀÇ ºÐº°·Â¿¡ ±â¹ÝÇÑ »õ·Î¿î ¿À·ù ºÐ¼® ¹æ¹ý·ÐÀ» Á¦¾ÈÇÑ´Ù. °¢ ´Ü°èº° ó¸® °úÁ¤¿¡¼ »ý¼ºµÈ »öÀξ ¹ÌÄ¡´Â ¿µÇâÀ» °è·®ÈÇÏ°í, ÀÌ °è·®È Á¤º¸¸¦ °¡½ÃÈÇÏ¿© ¿¬±¸ÀÚ/°³¹ßÀÚ°¡ °Ë»ö¿£ÁøÀÇ ¼º´ÉÀ» Çâ»ó ½Ãų ¼ö ÀÖ´Â µµ±¸¸¦ Á¦°øÇÑ´Ù.
|
¿µ¹®³»¿ë (English Abstract) |
For improving performance of search engines, iteration of analyzing the results of internal modules of search engines and then modifying their errors is required in general. However this job is highly labor-intensive and time-consuming for developers and researchers. Nevertheless, enhancing the methods of extracting good index terms from documents through this process is one of the most fundamental and important research topics. In general, the performance of search engines is enhanced by removing index terms that are negative to the performance or stressing on relatively important index terms. However, it is quite difficult for the researchers to investigate in detail and modify the problems occurred from multiple modules in a search engine. Therefore, we propose a failure analysis method based on both Bayesian inference network and discrimination power of a index term, in order for the researcher to easily analyze the effect of each index term on the search engine. To do this, we quantify the importance of each term and visualize them on the Bayesian inference network.
|
Å°¿öµå(Keyword) |
¿À·ù ºÐ¼®
»öÀÎ¾î °¡ÁßÄ¡
º£ÀÌÁö¾È Ã߷иÁ
Á¤º¸°Ë»ö
failure analysis
term weighting
bayesian inference network
information retrieval
|
ÆÄÀÏ÷ºÎ |
PDF ´Ù¿î·Îµå
|