大è¦æ¨¡ãƒ‡ãƒ¼ã‚¿ã§å˜èªžã®æ•°ã‚’æ•°ãˆã‚‹ - ny23ã®æ—¥è¨˜å¤§è¦æ¨¡ãƒ‡ãƒ¼ã‚¿ã‹ã‚‰ one-pass 㧠item（n-gram ãªã©ï¼‰ã®é »åº¦ã‚’æ•°ãˆã‚‹æ‰‹æ³•ã«é–¢ã™ã‚‹ãƒ¡ãƒ¢ï¼Žã“ã“数年，毎年ã®ã‚ˆã†ã«è¶…大è¦æ¨¡ãª n-gram ã®çµ±è¨ˆæƒ…å ±ã‚’ç©ºé–“ï¼æ™‚間効率良ã利用ã™ã‚‹ãŸã‚ã®æ‰‹æ³•ãŒæ案ã•ã‚Œã¦ã„る．最近ã ã¨ï¼Œ Storing the Web in Memory: Space Efficient Language Models with Constant Time Retrieval (EMNLP 2010) ã¨ã‹ï¼Žã“ã®è«–æ–‡ã§ã¯ï¼Œæœ€å°å®Œå…¨ãƒãƒƒã‚·ãƒ¥é–¢æ•°ã‚„ power-law を考慮ã—ãŸé »åº¦è¡¨ç¾ã®åœ§ç¸®ãªã©ï¼Œç´°ã‹ã„技術をä¸å¯§ã«çµ„ã¿ä¸Šã’ã¦ãŠã‚Šï¼Œã“ã‚Œãらã„工夫ãŒç´°ã‹ããªã£ã¦ãã‚‹ã¨log-frequency Bloom filter (ACL 2007) ãらã„ã‹ã‚‰ã‹ã‚‰å§‹ã¾ã£ãŸ n-gram é »åº¦æƒ…å ±ã®åœ§ç¸®ã®ç ”究もãã‚ãã‚åŽæŸã—ãŸã‹ã¨ã„ã†å°è±¡ï¼ˆã¡ã‚‡ã†ã©è«–文をèªã‚€ç›´å‰ã«ï¼Œã“ã®è«–æ–‡ã®7節ã®
{{#tags}}- {{label}}
{{/tags}}