( )

Size: px

Start display at page:

Download "( )"

えりかたかぎ
5 years ago
Views:

1 NAIST-IS-MT

2 ( )

3 , NAIST-IS- MT , i

4 ,, ii

5 Proposal and Evaluation of collaborative filtering method considering each weight of user s evaluation to restaurant Yoshihiro Kawaguchi Abstract Now, in widely the full-text search used, users become difficult to collect the information matched the preference of the user because they can t find the keyword matched the purpose or treat the flood of research results. So, Web service for satisfy user s special intention is appearing. This service provides the ordinary information that user request, but don t consider user s preference. It is difficult to treat the preference information which is generated from a huge amount of users. In this way, user s preference is unstable information for computer. That is why, when we search the restaurant information through the Internet, it is difficult to decide whether which information published is useful for me. So, the information filtering which considers user s preference is attracted. To treat the information of the preference, this method actualizes the providing information that we request by using content-base-filtering and collaborative filtering that it is technology of the information filtering. Inputting the number of the value explicitly is the information that is easy to apply to judge the user s preference and use this to filtering, we ll get the search result that matched the preference more. But, related works don t consider the real weight of the value that weighted by user because it uses the untouched value input numerically. So, in this research, Master s Thesis, Department of Information Systems, Graduate School of Information Science, Nara Institute of Science and Technology, NAIST-IS-MT , February 1, iii

6 determine the standard value by using the user distribution estimated against something, The standard value is the basing point whether it is good or not for user. It actualize to treat the real weight of the value by using the standard value. I suggest the information proposedsystem that considers the preference gap of each user and can match the preference of the user. As a result of implement and evaluate recommendation system utilizing collaborative filtering based on proposed method, error counts of proposed method is less than that of relative works. Keywords: recommendation,collaborative filtering iv

7 DB v

8 ( ) ( ) vi

9 vii

10 Web HP(Home Page) SNS(Social Networking Site) (Weblog) Amazon.com 1 TSUTAYA online 2 DVD Web 1 Amazon: 2 TSUTAYA online: 1

11 SNS (.com 3.com 4 ) SNS (.com) com: 4.com: 2

12 [1]

13 Web SNS Web 5 : 6 : 4

14 2.2 HP HP Web 5

16 (content-based-filtering) 1 A Web Web Web A A Web 7

17 1 8

18 Web Web 2 [2] [3] [4] [5] cold-start cold-start (cold-start condition) (collaborative filtering) 2 9

19 2 10

20 A C C 3 A A 3 A (memory-based method) (model-based method) [7] DB(Data Base) DB A B DB 1 DB 11

21 1 DB 1 GroupLens [1] GroupLens NetNews NetNews MovieLens GroupLens 1. DB 2. DB r AB = i(a i A)(B i B) i(a i A) 2 i(b i B) 2 (1) r AB : A B A i : A i A : A 12

22 B i : B i GroupLens ringo [6] CD r AB = i(a i 4)(B i 4) i(a i 4) 2 i(b i 4) 2 (2)

23 A a 3(5 ) A B (1) (2) 14

26 ((2.5*2)+(3*2)+(5*2)) /6= (0.5 5 )

27 A,B a n,b n 1.5(M(x) < x n ) a n, b n = 1(M(x) = x n ) (3) 0.5(M(x) > x n ) A,B :(x 1... x n ) A,B :M(x) A :(x 1... x n ) a n = x n M(x) (4) 18

28 A :M(x) 5 a n = x n M(x) + 5 (5) cos θ Cosine Similarity Cosine Similarity cosθ = a b a b (6) a, b: A,B

29 .com.com DB DB ( ) 5 ( ) ( 4 ) 20

30 4 21

31 5 22

32 DB(Data Base) 10 DB 5 DB

33 6 24

34 6 ( 6(a)) DB ( 6(b)) ( 6(c)) ( 6(d)) 25

35 DB HTML URL HTTP Web wget 7 HTML HTML HTML li class= name strong strong class name strong (.+?) strong 7 GNU Wget: 26

36 li class= address (.+?) strong IMG class=title ico alt= src= /images/ico rating food.gif IMG alt= src= /images/bar dtl 40.gif alt /bar dtl 40.gif /bar dtl (.+?).gif HP URL DB DB DB 2 ( 5) 3 ( 4) DB 27

37 2 (rst) rst id int ID rst name char(30) area1 char(10) area2 char(30) address text tel char(30) holiday char(30) hour text url text HP URL 3 (user) user id int ID user name char(30) passwd text 4 (rst value) id int ID user id int ID rst id int ID comment text value float 28

38 5 (criterion) user id int ID criterion1 float 1 criterion2 float 2 4 ( 4) DB DB 29

39 Cosine Similarity

40 Web CGI(Common Gateway Interface) 6 PHP MySQL CPU 6 AMD Opteron252*2 2.6GHz Main Memory OS Web 2GB DDR/400 Fedora Core5 Apache/2.2.2 PHP5.1.6 MySQL

41 7 8 9 ( ) 32

42 8 ( ) ( HP) 33

43 9 34

44 DB /

45 9 1 4 (3.5: ) A A 4 4 B 4 4 C D 3 3 E [1] [6] 1. DB 2. DB 3. 1 ( ) 4. 36

46 A 1 ( ) A 4 A A 4 A A B C D A B C D E A 4.5 ( ) 1 E A 1 A E 1 1 E ( 3,3.5,4) ( ) 10 ( ) 37

47 DB ( ) 4 4 0( ) 6.3 2( ) ( 2 )

49 6.4 3( )

50 11 41

51 7. (TSUTAYA Onlin) ( ) SNS 42

52 43

53 M1 LOWSAN LOWSAN

54 45

55 [1] Resnick, P., Iacovou, N., Suchak, M., Bergstrom, P., Riedl, J. GroupLens: An Open Architecture for Collaborative Filtering of Netnews. CSCW 94 Procedings. pp [2] Morita, M. and Shinoda, Y. Information Filtering Based on User Behavior Analysis and Best Match Text Retrieval. in Proc. of the 17th Annual International ACM-SIGIR Conference on Research and Development in Information retrieval. pp [3] Sakagami, H. and Kamba, T. Learning Personal Preference on Online Newspaper Articles from User Behaviors. Proc. of the Sixth International World Wide Web Conference, In Computer Networks and ISDN Systems, Vol. 29,pp [4],, Digital Reminder:, in Proc. of the 8th Workshop on Interactive Systems an Software (WISS 2000), pp [5],,,,,, DEWS2006, 2C-i [6] Shardanand, U. and Pattie Maes Social Information Filtering: Algorithm for Automationg Word of Mouth. CHI 95 Conference Procedings.pp [7].Vol.56 No.1,p

Web Web Web Web Web, i

22 Web Research of a Web search support system based on individual sensitivity 1135117 2011 2 14 Web Web Web Web Web, i Abstract Research of a Web search support system based on individual sensitivity