09한성희.hwp

(JBE Vol. 18, No. 1, January 2013) (Regular Paper) 181, 2013 1 (JBE Vol. 18, No. 1, January 2013) http://dx.doi.org/10.5909/jbe.2013.18.1.88 ISSN 2287-9137 (Online) ISSN 1226-7953 (Print) VOD TV a), a), a) Personalized TV Program Recommendation in VOD Service Platform Using Collaborative Filtering Sunghee Han a), Yeonhee Oh a), and Hee Jung Kim a).. VOD TV. TV,. TV VOD... Abstract Collaborative filtering(cf) for the personalized recommendation is a successful and popular method in recommender systems. But the mainly researched and implemented cases focus on dealing with independent items with explicit feedback by users. For the domain of TV program recommendation in VOD service platform, we need to consider the unique characteristic and constraints of the domain. In this paper, we studied on the way to convert the viewing history of each TV program episodes to the TV program preference by considering the series structure of TV program. The former is implicit for personalized preference, but the latter tells quite explicitly about the persistent preference. Collaborative filtering is done by the unit of series while data gathering and final recommendation is done by the unit of episodes. As a result, we modified CF to make it more suitable for the domain of TV program VOD recommendation. Our experimental study shows that it is more precise in performance, yet more compact in calculation compared to the plain CF approaches. It can be combined with other existing CF techniques as an algorithm module. Keywords : Recommender System, TV Program, VOD, Collaborative Filtering, Implicit Feedback

2 : VOD TV (Sunghee Han et al. : Personalized TV Program Recommendation in VOD Service Platform Using Collaborative Filtering). IP. VOD... KBS [1].. [3]. (Neighborhood Model) (Collaborative Filtering) (Latent Factor Model) [2]. (Explicit Rating). KBS VOD TV a) KBS (KBS Technical Research Institute) Corresponding Author : (Sunghee Han) E-mail: shhan9@kbs.co.kr Tel: +82-2-781-5232 Fax: +82-2-781-5299 Manuscript received October 31, 2012 Revised December 14, 2012 Accepted December 24, 2012. [ - ]. 1).,,.,.., VOD.,. (Implicit & Binary Data). KBS VOD KBS, K,. UI.. VOD TV 1). KBS.

(JBE Vol. 18, No. 1, January 2013). (Item Based CF : IBCF),.. (User Based CF : IBCF) [4]. 2 [2].,... [ 1] -. 1, 0.. (Missing Value) 0. [ 1] (De Ds)... TV TV. [5], (Confidence) [6]. TV. 1. - Fig. 1. User-Item matrix for CF and Episode-Series mapping

2 : VOD TV (Sunghee Han et al. : Personalized TV Program Recommendation in VOD Service Platform Using Collaborative Filtering) 2. 2 '' Fig. 2. Distribution of 'number of episodes per a program' for 2 months in Conting platform.. [ 2] KBS 2) 2010 3~4 2 VOD. VOD. 2 67 1067 1, 86 16. TV,.... VOD TV 1.. [ 1] r(u,i) rp(u,i). [ 1],, Du, De, Ds. (1). 1. Table 1. User's consumption history data & program information to draw the user's preference for program u i u i u i u i 2) http://conting.conpia.com

(JBE Vol. 18, No. 1, January 2013) (1) (1). log.).,. () 2., i f i f [ 2]. (λ, [ 3] 2 2 3. Fig. 3. Weight for conversion to program rating Du Ds ( - ) Rp. Rp 0 1 0 (Tanimoto) [7]. [2] (Cold Start Problem) [8]. [4]. (Cosine Similarity : COS)(Pear- son Correlation Coefficient : PCC). (4)(5). 0.

2 : VOD TV (Sunghee Han et al. : Personalized TV Program Recommendation in VOD Service Platform Using Collaborative Filtering) Rp De Ds (Computa- tional Complexity). 2..1 i j.. [4]. (6) N. i f TV. 4. Fig. 4. Contents arrangement by episode unit in Conting.. KBS VOD [ 4]. (6) (6) (7). (6) (6) (7). i f (6),(7).. TV. [ 5]. [ 5].. C,.

(JBE Vol. 18, No. 1, January 2013) 5. - Fig. 5. Mapping from program to program episode based round-robin rule C.,. [ 5]. (Computational Complexity)... (Training Set) (Test Set) KBS VOD. KBS 2010 3 4, 5. [ 2]..1 [ 3]. 2. Table 2. Dataset specification () () () 61545 1067 312298 2010 3~4 61545 67 89408 (2) 10000 1379 33060 2010 5 (1). 1.., 3. ' Table 3. Data conversion example for a user ID ( 9 ) A-1, A-2, A-3, A-4 B-1, B-2, B-3, B-4 C-2 A : (1~5) log = 0.66 log B : (1~40) = 0.44 C : (1~2) log = 0.25

2 : VOD TV (Sunghee Han et al. : Personalized TV Program Recommendation in VOD Service Platform Using Collaborative Filtering). MAE(Mean Absolute Error) (Classification Accuracy Metrics) F1. F1 (Precision) (Recall)(Harmonic Mean),. 2. [ 6] 10000. F1. RoSE (Recommendation on Series- Episode). (PCC, COS). IBCF(Item-Based Collaborative Filtering) RoSE. RoSE, (2), (3). RoSE. COS RoSE 3.65. RoSE COS PCC. RoSE PCC 10.6%. 6. Fig. 6. Evaluation result for recommendation

(JBE Vol. 18, No. 1, January 2013).. [ 7] PCC 10000. RoSE () 4.4%. 7. Fig. 7. Run-time for generating recommendation. TV. TV. (latent).. TV.,,.,.... [1] Soo-Young Oh, Yeonhee Oh, Sunghee Han, Hee Jung Kim, Broadcast Content Recommender System based on User s Viewing History, JBE, Vol. 17, No. 1, pp130~140, Jan, 2012 [2] Xiaoyuan Su and Taghi M. Khoshgoftaar. "A Survey of Collaborative Filtering Techniques," Advances in Artificial Intelligence Vol. 2009, Article No. 4, 2009. [3] István Pilászy and Domonkos Tikk. 2009. "Recommending new movies: Even a few ratings are more valuable than metadata," in Proc. Recsys 2009, ACM, New York, 2009. [4] Badrul Sarwar, George Karypis, Joseph Konstan, and John Riedl. "Item based collaborative filtering recommendation algorithms," in Proc. 10th International Conference on WWW, ACM, NewYork, 2001. [5] Eunhui Kim, Shinjee Pyo, Eunkyung Park, and Munchrul Kim. "An automatic recommendation scheme of TV program contents for IPTV

한성희 외 인 서비스 플랫폼에서 협력 필터링을 이용한 프로그램 개인화 추천 2 : VOD TV (Sunghee Han et al. : Personalized TV Program Recommendation in VOD Service Platform Using Collaborative Filtering) [6] [7] personalization," IEEE Transactions on Broadcasting, Vol. 57, No. 3, 2011. Yifan Hu, Yehuda Koren and Chris Volinsky. "Collaborative filtering for implicit feedback datasets," in Proc. 8th IEEE International Conference on Data Mining, pp. 263-272, 2008. Manzhao Bu, Shijian Luo, and Ji he. "A fast collaborative filtering al- gorithm for implicit binary data," IEEE 10th International Conference on Computer-Aided Industrial Design & Conceptual Design, pp. 973 976, 2009. Hyung Jun Ahn. "A new similarity measure for collaborative filtering to alleviate the new user cold-starting problem," Information Sciences 178, ScienceDirect, pp. 37-51, 2008. [8] 저자소개 한성희 - 년 2월 : 고려대학교 전기전자전파공학부 학사 년 2월 : 고려대학교 전기공학과 석사 년 1월 ~ 2006년 5월 : 삼성전자 무선사업부 선임연구원 년 3월 ~ 현재 : KBS 기술연구소 연구원 주관심분야 : 콘텐츠 추천 시스템, 방송 자막 활용, 하이브리드 방송 플랫폼 2001 2003 2003 2007 오연희 - 년 2월 : 서울대학교 컴퓨터공학과 학사 년 9월 : University College London, MSc in DCNDS 석사 년 11월 ~ 2009년 10월 : NHK 기술연구소 객원연구원 년 1월 ~ 현재 : KBS 기술연구소 선임연구원 주관심분야 : 콘텐츠 추천/검색, 메타데이터, 정보 추출, 멀티미디어 콘텐츠 서비스 2000 2002 2008 2003 김희정 년 2월 : 이화여자대학교 전자계산학과 학사 년 2월 : KAIST 전산학과 석사 년 ~ 현재 : KBS 기술연구소 방송기술연구부장 주관심분야 : 콘텐츠 추천, 동영상 편집전송, 컴퓨터 그래픽스 - 1985-1988 - 1988-97