Vol. 47 No. 4 Apr. 2006 y1 y2;y3 y1 y1 y1 y4;y1 - - SVM User Modeling based on Location History YUTAKA MATSUO, y1 NAOAKI OKAZAKI, y2;y3 YOSHIYUKI NAKAMURA, y1 TAKUICHI NISHIMURA, y1 KÔITI HASIDA y1 and HIDEYUKI NAKASHIMA y1 Recent development for location detection techniques enables us to obtain location histories for users in a ubiquitous environment. This paper describes a new method to infer user attributes from a user s location history. Using the number of counts each sensor detects each user, we can obtain a user-sensor matrix, which is similar to document-term matrix in the context of information retrieval. The problem to detect a user s attributes can be reduced into text classification problem, to which support vector machine can be effectively applied. We also propose a method to measure the importance of sensors. 1. 1)ο3) 4) Active Badge 5) Active Badge GPS RFID y1 National Institute of Advanced Industrial Science and Technology y2 University of Tokyo y3 University of Manchester y4 Future University, Hakodate 6) 7) 6),8) 9),10) 11)ο13) Heckmann 13) A user model is a knowledge source in a system which containes explicit assumptions on all aspects of the user that may be relevant to the behavior of the system. These assumptions must separable by the system from the rest of the systems s knowledge. 1
2 Apr. 2006 13) Heckmann attribute-value 2004 ffl ffl ffl Workshop on User Modeling for Ubiquitous Computing (2003) Workshop on Personalized Context Modeling and Management for UbiComp Applications(2005) UbiqUM2006(Workshop on Ubiquitous User Modeling) 2005 Journal of User Modeling and User-Adapted Interaction User Modeling in Ubiquitous Computing 3 4 5 6 7 2. 14),15) Kobsa 11)
Vol. 47 No. 4 3 Wilson 7),16) Narrotor 17) GPS location Hightower GPS WiFi LAN) GSM1 18) 3. 3.1 CoBIT ID CoBIT 19),20) CoBIT ID ID CoBIT 1 3-5m ID 2 ID CoBIT 3 ID CoBIT 2003 2004 21),22) 3 ID ID IC RDIF GPS 17) RFID 3.2 2004 2 16 20 94 4F CoBIT 4 1000m 2 3 1 66 CoBIT 47 23) ID
4 Apr. 2006 Sensor ID User ID Date/Time 1 2 2004-02-27 09:24:45 3 2 2004-02-27 09:25:12 3 2 2004-02-27 09:25:22 4 4 2004-02-27 10:01:34... 3 4 2 1 CoBIT 170 24317 20273 431.3 23.8
Vol. 47 No. 4 5 4. 4.1 ID - - W n m n m W ij s j u i 4 3 W = 0 B @ 1 2 2 4 1 0 2 0 3 2 0 0 1 C A (1) f g 2 1 0 3 1, 0, 1 u1 u2 u3 s1 s2 s3 s4 coffee 1 2 2 4 1 1 0 2 0 0 3 2 0 0 1 u1 u3 s1 s4 coffee s1 s4 coffee 1 0 u4 coffee 1 0 s1 s2 s3 s4 coffee u4 1 2 0 0? s2 s2 coffee 1 u4 coffee 1 - W n m - 24) Support Vector Machine (SVM) SVM 25) RBF radius basis function SVM 26) 27) - 7 4.2 tf idf tf idf
6 Apr. 2006 1 25, 24-29, 30-34, 35-39,40 SC,,, SC A, B, C, D,,,,,,, A, B, C, D, E, F A, B 2 u i s j tfidf (s j;u i)=freq(sj;u i) idf (sj ) (2) freq(s j;u i) sj u i idf (s j ) idf (s j ) = log(n=uf (sj )) (3) n uf (s j ) sj uf (s j) idf (s j ) 8 u i s j a ij ffl a ij = freq(sj;u i) ffl a ij = ( 1 if freq(s j;u i) thre 0 otherwise thre 1 ffl IDF ( idf (s j) if freq(sj;u i) thre a ij = 0 otherwise ffl ffl TFIDF a ij = tfidf (sj;u i) 1 m a P ij m (4) i=1 aij a normalized ij = 5. 3.2 2 F (%) Recall(%) Precision(%) 44.45 73.56 37.75 43.92 65.83 41.62 TFIDF 44.28 71.62 37.38 IDF 44.37 68.45 45.33 54.46 68.83 49.23 40.73 63.80 40.97 IDF 41.23 61.02 41.46 TFIDF 53.00 65.50 47.88 5.1 1 SVM 2 SVM Leave-one-out RBF 2 8 Recall Precision F F Recall Precision F = 2Recall Precision Recall + Precision Recall 73.56 73.56% Precision 37.75 37.75% Recall 70% Precision 50% F
Vol. 47 No. 4 7 3 F F (BL) ( ) 41.79 (32.95) age0* 63.15 (25.53) 25 age1 17.39 (25.53) 24-29 age2 48.49 (45.28) 30-34 age3 19.05 (29.17) 35-39 age4* 60.87 (39.22) 40 ( ) 41.33 (32.66) position0 13.33 (20.00) SC position1 41.03 (44.83) position2* 60.61 (42.11) position3 25.00 (36.36) position4* 66.67 (20.00) ( ) 59.18 (28.50) team0* 58.83 (34.04) A team1* 60.00 (30.43) B team2 21.05 (22.73) C team3* 46.15 (26.67) D team4* 83.33 (26.67) team5* 85.71 (30.43) ( ) 48.75 (49.42) atd0 50.00 (60.00) atd1 48.65 (51.06) atd2* 47.62 (37.21) ( ) 57.22 (48.90) coffee0 58.83 (56.41) coffee1 66.67 (60.00) coffee2* 46.15 (30.30) ( ) 64.13 (57.58) smoking0 91.89 (93.33) smoking1* 36.36 (22.22) ( ) 67.00 (28.29) room0* 72.73 (23.08) B room1* 77.78 (26.42) C room2* 66.67 (19.61) D room3* 62.50 (29.63) A room4* 46.15 (29.63) E room5* 76.19 (41.38) F 61.86 (63.74) station0 83.72 (83.72) A station1 40.00 (43.75) B 54.46 (38.38) F TFIDF 10 5.2 F 3 F F smoking0 F BLF 10 * 25 40 F 60 F A B) 1 F
8 Apr. 2006 F CoBIT 70% E SVM 5 9 6. 6.1 1 P n ffl freq w(s j)= j=1 freq(sj;ui) ffl user-freq w(s j)=uf (sj ) ffl ffl P n TFIDF tfidf w(s j)= j=1 tfidf (sj;ui) 3 nornalized (4) nx w(s j )= j=1 a normalized ij ffl wfreq nx w(s j )= freq(s j;u i) log(m=sf (ui)) j=1 sf (u i) ui TFIDF 7
Vol. 47 No. 4 9 F value 60 55 50 45 40 35 wfreq tfidf freq 30 user-freq tfidf(normalized) 25 freq(normalized) user-freq(normalized) random 20 0 5 10 15 20 25 30 5 Number of enabled sensors F 6.2 5 SVM 3 * 18 * F 18 28),29) F F (random) F (wfreq) (freq (user-freq) TFIDF (tfidf) 5 7. 1 A-B-C A-C A-B-C - ffl ffl bag of words ffl ffl ffl
10 Apr. 2006 3 10 2005 8. - SVM tf idf 1) (2002). 2) (2004). 3) Vol.18, No.4, pp. 48 56 (2001). 4) Hightower, J. and Borriello, G.: Location Systems for Ubiquitous Computing, IEEE Computer, Vol.34, No.8, pp.57 66 (2001). 5) Want, R., Hopper, A., Falcao, V. and Gibbons, J.: The Active Badge Location System, ACM Transactions on Information Systems, Vol.10, No.1, pp.91 102 (1992). 6) Vol.20, No.5, pp.579 586 (2005). 7) Wilson, D. H.: The Narrator : A Daily Activity Summarizer Using Simple Sensors in an Instrumented Environment, Proc. UbiComp 2003 (2003). 8) Vol.46, No.12, pp.1373 1381 (2005). 9) (2005). 10) -CONSORTS Vol.47, No.2, pp.105 108 (2006). 11) Kobsa, A.: Generic User Modeling Systems, User Modeling and User-Adapted Interaction, Vol.11, pp. 49 63 (2001). 12) Brusilovsky, P.: Methods and techniques of adaptive hypermedia, User Modeling and User-Adapted Interaction, Vol.6, pp.87 129 (1996).
Vol. 47 No. 4 11 13) Heckmann, D.: Ubiquitous Use Modeling, Ph.d thesis, University of Saarland (2005). 14) No.2003-UBI-002, pp.61 66 (2003). 15) (2001). 16) Wilson, D., Long, A. and Atkeson, C.: A ContextAware Recognition Survey for Data Collection Using Ubiquitous Sensors in the Home, Proc. CHI 2005 (2005). 17) Ashbrook, D. and Starner, T.: Using GPS to learn significant locations and predict movement across multiple users, Personal and Ubiquitous Computing, Vol.7, No.5, pp.275 286 (2003). 18) Hightower, J., Consolvo, S., LaMarca, A., Smith, I. and Hughes, J.: Learning and Recognizing the Places We Go, Proc. UbiComp 2005 (2005). 19) CHOBIT Vol.44, No.11, pp.2670 2680 (2003). 20) Nakamura, Y., Nishimura, T., Itoh, H. and Nakashima, H.: ID-CoBIT: A Battery-less Information Terminal with Data Upload Capability, Proc. IECON 2003 (2003). 21) 2003 Vol.19, No.1, pp.43 51 (2004). 22) Nishimura, T., Nakamura, Y., Itoh, H. and Nakamura, H.: System Design of Event Space Information Support Utilizing CoBITs, Proc. ICDCS 2004, pp.384 387 (2004). 23) Schulz, D., Fox, D. and Hightower, J.: People Tracking with Anonymous and ID-Sensors Using Rao-Blackwellised Particle Filters, Proc. IJCAI-03, pp.921 928 (2003). 24) Manning, C. and Schütze, H.: Foundations of statistical natural language processing, The MIT Press, London (2002). 25) Vapnik, V.: The Nature of Statistical Learning Theory, Springer-Verlag (1995). 26) ( ) (2005). 27) - - Vol.42, No.7, pp.676 683 (2001). 28) Joachims, T.: Text categorization with support vector machines, Proc. ECML 98, pp.137 142 (1998). 29) Mladenic, D., Brank, J., Grobelnik, M. and Milic- Frayling, N.: Feature selection using linear classifier weights: interaction with classification models, Proc. SIGIR 2004, pp.234 241 (2004). ( 16 11 28 ) ( 17 2 4 ) 1997 2002 2005 10 Web AAAI INSNA 2001 2003 2003 2005 (National Centre for Text Mining) 2006 2007 1994 1996 1997 2006 IEEE
12 Apr. 2006 1992 () NKK( ) X 1995 RWCP 1998 NKK( ) 1999 RWCP 2001 2005 ( ) ( ) ACM 1983 2001 2004 1981 1986 1988 1992