Recsys2013èªã¿ä¼ã§ Personalized next-song recommendation in online karaokes 㨠Nonlinear latent factorization by embedding multiple user interests ãèªã¿ã¾ãã
ã¨ãããã㧠Gunosy æ°ãªãã£ã¹ã§è¡ããã RecSys 2013 (Hong Kong) – RecSys ã®è«æèªã¿ä¼ã«åå ããï¼çºè¡¨è
ã®çæ§ãç²ãæ§ã§ããï¼
èªå㯠Personalized next-song recommendation in online karaokes 㨠Nonlinear latent factorization by embedding multiple user interestsã® short paper ã2æ¬èªãã ï¼
Personalized next-song recommendation in online karaokes
Personalized next-song recommendation in online karaokes(pdf)
ãªã³ã©ã¤ã³ã®ã«ã©ãªã±ãµã¼ãã¹ã§æ¬¡ã«ä½ãæãããäºæ¸¬ããè«æï¼ããæ²ãæã£ã次ã«ã©ã®æ²ãæããã®ãã«ã³ãæ§ï¼åã³ããã¦ã¼ã¶ãã©ã®æ²ãæããã®é¢ä¿æ§ã Metric Enbedding ã§å¦ç¿ï¼çµæãè¦ãæãï¼ã¦ã¼ã¶ãã¨ã«æ§ç¯ããä¸æ¬¡ã®ãã«ã³ãé£éã¨ãããªã«ç²¾åº¦ãå¤ãã£ã¦ããªãã®ãæ¸ãï¼
Nonlinear latent factorization by embedding multiple user interests
Nonlinear latent factorization by embedding multiple user interests(pdf)
ã¦ã¼ã¶ã¨ã¢ã¤ãã ã§æ§æãããè¡åãè¡åå解ããæï¼é常ã¯ä½æ¬¡å ã«è½ã¨ãããã ãã©ï¼å¤§ããªä»®å®ã¨ãã¦ãã¦ã¼ã¶ã¯måã®èå³ãæã£ã¦ãã¦ï¼måã®èå³ããã¢ã¤ãã ãçæããããã¨ãããã®ãç½®ãããï¼ããã©ãï¼ã¢ã¤ãã ãå¤å²ã«æ¸¡ãå ´åã§ã¯ãã®ä»®å®ã¯å³ãã(é£åã¨æ¬ã®å好ãä¸ã¤ã®ãããã¯ã§è¡¨ç¾ã§ããã?)ï¼ãªã®ã§ï¼ã¢ã¤ãã å´ã¯m次å ã§æã£ã¦ï¼ã¦ã¼ã¶å´ã¯ãã³ã½ã«ã§è¡¨ç¾ããï¼ã¹ãã¼ã¹ã«ãªããããã®ã§ã¯ãªããã¨ããã®ãçåï¼
è©ä¾¡ã¢ã¤ãã ã¨éè©ä¾¡ã¢ã¤ãã ã®ã©ã³ã¯å·®ãèããæ失é¢æ°ï¼ãã¾ãè¦ããã¨ã¯ãªãã£ããã©çµæ§ãããããããåå¼·ã«ãªã£ãï¼
ãã®ä»ã®è«æã¯ä»¥ä¸ã«ï¼
Personalized News Recommendation with Context Trees
Personalized News Recommendation with Context Trees(pdf)
- ãã¥ã¼ã¹æ¨è¦ã®è«æ
- ãã¥ã¼ã¹æ¨è¦ã«æ±ããããè¦ä»¶
- æ°ããè¨äºã«ä¾¡å¤ããã
- æ°ããè¨äºãã©ãã©ã追å ãããã®ã§ï¼ã¤ã³ã¯ãªã¡ã³ã¿ã«ã«å¦çãããå¿ è¦ããã
- ææ¡ææ³: context trees
- ãã«ã³ãã¢ãã«ã«è¿ã(éå»ã®kåã®è¨äºãã次ã®è¨äºãé¸ã¶)
- variable-order Markov model
- 人çã²ã¼ã ã¿ããã«ï¼è¨äºããã¼ãã§æã¤ããªã¼ããã£ã¦ï¼ã¦ã¼ã¶ãè¨äºãèªããã³ãã¼ãã移åãã
- è¨äºãã¨ã® topic ãkd treeã§åå²ãã¦æ ¼ç´
- ãã«ã³ãã¢ãã«ã«è¿ã(éå»ã®kåã®è¨äºãã次ã®è¨äºãé¸ã¶)
- expert model
- ãã¼ããã¨ã«æã¤äºæ¸¬ã¢ãã«
- éã¿ã¯æ¨è¦ãæåãããã³æ´æ°ãã
- expert model ã¯ä»¥ä¸ã®3ã¤ã®éã¿ä»ãå
- standard model
- é »åº¦ãã¼ã¹ã§ã©ãã ãèªã¾ããã
- popular
- ç´è¿ã§èªã¾ãã¦ãæ°ã®é »åº¦
- freshness model
- 誰ã«ãèªã¾ãã¦ãªãææ°è¨äºF件ã使ã
- standard model
- å®é¨
- 人æ°è¨äºãå«ããã¨ï¼ææ¡ææ³ã¯ç²¾åº¦ãè¯ããªã
- 人æ°è¨äºãæãã¨ç²¾åº¦ãè¯ã
- èè ãããããéè¦ãªã®ã¯äººæ°è¨äºä»¥å¤ã®ææ¨ã ã
Query-Driven Context Aware Recommendation
Query-driven context aware recommendation
- ã¦ã¼ã¶ï¼ã¢ã¤ãã ï¼ã³ã³ããã¹ãã使ã£ã¦ LDA
- ææ¡ã¢ãã«: ãããã¯æ··åæ¯çï¼ãããã¯å¥ã¢ã¤ãã åå¸ï¼ãããã¯å¥ç¹å¾´éåå¸(ã³ã³ããã¹ã)ãå¦ç¿
- ã¢ã¤ãã ã®èæ¯ã«ããã³ã³ããã¹ããã¢ãã«ã«å°å ¥
- ã°ã©ãã£ã«ã«ã¢ãã«çã«è¨ãã°ï¼ãããã¯ããã¢ã¤ãã ã¨ç¹å¾´éãçæãããæã
- ã¢ã¤ãã¢: context aware
- ã¢ã¤ãã ã«å¯¾ãã¦å±æ§ã¨ã¹ã³ã¢ãã¤ãã¦ããæã
- ãã®ç¹å¾´éãã¢ã¤ãã ã¨åããããã¯ãæã¤
- è¨ç®ãããã®ã¯ p(ã¢ã¤ãã |ã³ã³ããã¹ãï¼ã¦ã¼ã¶)
- p(s|q, p; theta, mu, phi) = p(q|s, p; theta, mu, phi) * p(s | p, theta, mu)
- å解ã®éä¸ã§ç¹å¾´éãç¬ç«ã§ããã¨ä»®å®
- ãã¢ã¤ãã ã®ããããã¨ãç¹å¾´éã®ãããããç©ã§è¡¨ã
- çµæ: ãã¥ã¼ã¹è¨äºã ã¨kNNã«åã¦ãªãï¼é³æ¥½æ¨è¦ã ã¨æ¹å
- ç¹å¾´éã¨ãããã¯ãç°ãªã£ã¦ããã®ã§æ¹åãã?
A Fast Parallel SGD for Matrix Factorization in Shared Memory Systems
A Fast Parallel SGD for Matrix Factorization in Shared Memory Systems(pdf)
- best paper
- è¡åå解ã®SGDãé«éã«ä¸¦åè¨ç®ããã
- hogwild: ãªãã§ãããããã©ã³ãã ã«æ´æ°ãã¾ããï¼è¡çªãèµ·ãããã©æ°ã«ãªãï¼
- DSGD: é£è»ã絶対éãªããªãããã«é ç½®ããããªã§è¡åãåå²
- DSGDã®åé¡: å ¨ãã¼ãã®è¨ç®å¾ ã¡ã³ã¹ãï¼åãã¼ãã«ãããå¦çæéã®å·®ãããã«ããã¯ã«ï¼
- ææ¡ææ³: è¡åããã¼ãæ°ãããç´°ããåå²ï¼è¨ç®ãçµãã£ããã¼ãã«ã©ãã©ãæ¯ã£ã¦ãããã¨ã§è¨ç®å¾
ã¡ãã¼ããçããããªã
- ç´°ããªãã¥ã¼ãã³ã°ã¨ãã¦ãããã¯åä½ã§ã¯ã©ã³ãã ã¢ã¯ã»ã¹ï¼ãããã¯å ã¯ã·ã¼ã±ã³ã·ã£ã«ãªã¢ã¯ã»ã¹ã«ãã¦ãã£ãã·ã¥ãã¹ãæ¸ãã
ææ³
ã©ã®è«æã(æå¾ã®parallel SGD以å¤ã¯)ææ³ã ãã§ãªãï¼ãã©ããã£ã¦ææ¡ææ³ã®ç²¾åº¦ãã¯ããããã¨ããç¹ãåé¡ã«ãªã£ã¦ããããã«æããï¼
ä»åç´¹ä»ããè«æã«ããã¦ï¼æ£ç´ãªã¨ããï¼æã人æ°ãªååãè¦ãã¦ããã°é£ããã¢ã«ã´ãªãºã ã使ããã¨ã大æµããããã®ç²¾åº¦ãåºã¦ãã¾ã£ã¦ããï¼ãããã¦ã¼ã¶ã«ã¨ã£ã¦ãããè¯ããã¨ããã¨å®éã¯ããã§ã¯ãªãã¦ï¼æå¤æ§ã ã£ããï¼ã»ã¬ã³ãã£ããã£ã ã£ããï¼ä»ã®è¦ç´ ã§ã¦ã¼ã¶ããã®ãµã¼ãã¹ã«æºè¶³ãããå¦ãï¼ã¨ããç¹ã決ã¾ã£ã¦ããã¨èããããï¼
ãããå®é¨ã§ã©ãã¯ãããï¼ã©ã主張ããããéè¦ã§ããããã«æããï¼