2008-12-01ãã1ã¶æéã®è¨äºä¸è¦§
ICML2008ã§çºè¡¨ãããDredzeãã®Confidence Weighted Linear Classificationãèªãã ãããã¯ç·å½¢åé¡å¨ãå¦ç¿ããæ°ãããªã³ã©ã¤ã³å¦ç¿åã¢ã«ã´ãªãºã ã®ææ¡ã§ãããããã«ä½¿ããå®è£ ã¨ãã¦ã¯OLLã¨ãããªã¼ãã³ã½ã¼ã¹ã®ã©ã¤ãã©ãªããããå®éã«è¯ãå®é¨çµâ¦
Rubyã§ä¹ ã ã«æ¸ãæ¨ã¦ã¹ã¯ãªãããæ¸ãã¦ããã ãã©ãæ¨æºå ¥åããä¸è¡ä¸è¡ãã¡ã¤ã«åãèªã¿è¾¼ãã§ãã¡ã¤ã«ãå¦çãããã¨ããç°¡åãªããã°ã©ã ãæ¸ãã¦ãã¦ããã¡ã¤ã«ãéããã«å°ã£ããããããæ©ãã æå¥ãçµå±ã¯æ¨æºå ¥åããèªã¿è¾¼ãã ãã¡ã¤ã«åã«æ¹è¡â¦
Complement Naive BayesãSVMããéããã¼ã¨ä¸»å¼µãã¦ããã®ã§ãSVMããªããæè¿ã¯éããªã£ã¦ããããããã¨ããäºãç´¹ä»ãã¦ã¿ãããè¿å¹´ã¯SVMãªã©ã®å¦ç¿ãé«éã«è¡ãã¨ããææ¡ãè¡ããã¦ãããå®è£ ãå ¬éããã¦ãããã®ãããããã®ä¸ã®ä¸ã¤ã«liblinearã¨â¦
æ°ã¯ã¦ãæ£å¼ãªãªã¼ã¹è¨å¿µã¨ãããã¨ã§ããããªãªã¼ã¹ããä½é±éãçµã£ã¡ãã£ããã©ã æ°ã¯ã¦ãªããã¯ãã¼ã¯ã§ã¯ããã¯ãã¼ã¯ã¨ã³ããªãã«ãã´ãªã¸ã¨èªåã§åé¡ãã¦ãããããã®ã«ãã´ãªåé¡ã«ä½¿ããã¦ããã¢ã«ã´ãªãºã ã¯Complement Naive Bayesãããã仿¥â¦
ãã¿ã¼ã³èªèã¨å¦ç¿ã®çµ±è¨å¦ã®p.153ã«ããã®ã³ã°ã¯ä¸¦ååãå¯è½ã§ããã®ã§ã並åè¨ç®æ©ãç¨ãããã¨ã«ãã£ã¦è¨ç®æéãå¤§å¹ ã«ç縮ã§ãããããã¼ã¹ãã£ã³ã°ã¯é次çã«ããè¨ç®ã§ããªãã®ã§ãè¨ç®æéã¯çæããå¼±ä»®èª¬ã®æ°ã«æ¯ä¾ãã¦é·ããªããã¨ã«ãªããã¨æ¸â¦