Search
Search

연구성과물 검색 타이틀 이미지

HOME ICON HOME > Search by Achievements Type > Reports View

Reports Detailed Information

https://www.krm.or.kr/krmts/link.html?dbGubun=SD&m201_id=10011448&local_id=10014346
유비쿼터스 환경에서 개체간의 자율적 협업에 기반한 추천방법 개발
Reports NRF is supported by Research Projects( 유비쿼터스 환경에서 개체간의 자율적 협업에 기반한 추천방법 개발 | 2005 Year 신청요강 다운로드 PDF다운로드 | 김재경(경희대학교) ) data is submitted to the NRF Project Results
Researcher who has been awarded a research grant by Humanities and Social Studies Support Program of NRF has to submit an end product within 6 months(* depend on the form of business)
사업별 신청요강보기
  • Researchers have entered the information directly to the NRF of Korea research support system
Project Number B00200
Year(selected) 2005 Year
the present condition of Project 종료
State of proposition 재단승인
Completion Date 2007년 05월 31일
Year type 결과보고
Year(final report) 2007년
Research Summary
  • Korean
  • 본 연구에서는 유비쿼터스 환경에서 개인화 서비스 구현을 위한 고객간 추천네트워크로써, Customer-net을 제안한다. Customer-net은 고객 단말기에 장착된 사용자 모델을 기반으로 개인 정보의 저장 및 관리가 이루어지도록 함으로써 개인의 정보관리 권한을 강화하였다. 또한, 추천절차에서 네트워크를 구성하고 있는 최근접 고객의 정보만을 활용함으로써 개인단말의 처리속도의 한계를 극복하고자 하였으며, 최근에 발생된 구매데이터를 기반으로 추천목록을 생성하여 고객의 선호변화를 동적으로 반영하고자 하였다.
    Customer-net에서 customer model은 다음 네가지 부분으로 구성되어 있다—personal information set, item information set, recommender set, 그리고recommendee set. 그리고 Customer-net은 상호 협업적인 다음 4개의 agent로 구성되어 있다 - information dissemination agent, personal information management agent, information retrieval agent, 그리고 customer network formation agent
    Customer-net의 성능을 실험하기 위하여, 멀티미디어 컨텐트 구매 데이터를 이용한 비교실험을 하였다. 추천의 정확성과 추천목록 산출속도에서 일반적인 협업필터링 및 베스트셀러 추천방법에 비하여Customer-net의 월등한 성능을 확인하였을 뿐만 아니라, 추천의 질에서도 뛰어남을 실험으로 확인하였다. 또한 CF에서 네이버를 선정할 때, 많은 구매 기록을 가지고 있는 고객만을 대상으로 했을 때가 가장 좋은 정확도를 보였다. 이와 같은 결과는 현재의 제한된 정보만을 이용하면서도, 즉 전체 고객 정보를 다 이용하지 않고 로컬 정보만을 이용하면서도 뛰어난 결과를 보여 줌으로서, 유비쿼터스 환경에서의 매장에서 수시로 바뀌는 고객들을 로컬 네트워크를 구성하여 뛰어난 추천을 할 수 있다는 것을 제시한다. 즉, Customer-net이 향후 유비쿼터스 환경에서 개인화 서비스 제공을 위한 현실적인 추천 방법임을 검증하였다.
  • English
  • Customer-net is operated in a community of devices, customers, and services that cooperate together to achieve common goals. Our definition of the community differs from the traditional ones restricted by a geographic area or a user group. In out study, all the interaction and collaboration for recommendation is assumed to take place within the Customer-net space. Therefore, the customers connected to Customer-net require functions to manage and maintain their own client-side customer model. Customer-net is processed with a model-based learning algorithm to manipulate information on the behalf of each customer. The customer model for Customer-net consists of four parts—personal information set, item information set, recommender set, and recommendee set—and works based on local information rather than information from all references in the space. Customer-net has four agents—information dissemination agent, personal information management agent, information retrieval agent, and customer network formation agent.
    Experimental results support our view that Customer-net is a realistic solution for ubiquitous personalization services. Customer-net worked dramatically better than other comparative systems. These results mean that an active use of less but current information improves recommendation performance in the ubiquitous environment. And it shows that using large amount of purchase information worked better than using medium and small amount. This implies that selecting customers with large amount of purchase information results a better accuracy when deciding neighbors in the CF procedure.
    Customer-net makes it possible for customers to buy products with much less effort in information search and much lower connection time. Customer-net can provide retailers in a physical space with networking platform consisting of customers who can self-manage their information with their personal devices.
Research result report
  • Abstract
  • This paper proposes a Customer-net, a local customer network in ubiquitous shopping spaces, to provide recommendation service. The Customer-net works on customers’ portable devices which encapsulate client-side customer model. Customer-net essentially follows the ground principle of CF, but it differs in that client-side recommendation for the ubiquitous computing environment is achieved. While Customer-net prevents vendors from gathering all customer information, it provides individual customers with the same or similar recommendation services that traditional CF systems do. Recommendation on the client-side is an approach to empowering customers with more control over their personal information. Recommendation on the client-side has been suggested to empower customers with more control over their personal information. But the limitation of the computing ability and storage of personal device makes it necessary to lessen the burden of customers’ personal devices, as they encapsulate client-side customer models. Customer-net attempts to improve recommendation accuracy with less computational time by focusing on local relationship of customers. Moreover, Customer-net suggests more active customers into their neighbors in CF procedure to increase the accuracy of recommendation with less effort. We implemented such customer networks in the area of multimedia content recommendation and validated that our approach attempts to improve recommendation accuracy with less computational burden by focusing on local and powerful relationship of customers. This proves that Customer-net has a good potential to be a realistic solution to the problems encountered in personalization service in ubiquitous environment.
  • Research result and Utilization method
  • Introduced is Customer-net, a network of customers for effective and efficient recommendations in a ubiquitous environment. Customer-net is constructed based on collaborative filtering algorithm and local learning. The recommendations made by Customer-net are locally processed by the personal recommender agent. The agents operate the network based on current purchase information in order to reflect preference changes of customers into the processes of network formation and recommendation generation.
    Each personal recommender agent is composed of an information dissemination agent, personal information management agent, information retrieval agent, and customer network formation agent. Customer-net has the following characteristics; (1) Customer-net deals with the most current preference of customers to adapt preference change in recommendations, (2) Customer-net learns customer’s preference in real time from customer’s content selections without requiring customer’s explicit ratings, (3) customer’s event, such as saving an image, triggers recommendations with push way without centralized control. (4) Similar neighbor customers are dynamically selected from neighbor customer’s neighbors only. (5) The customers with large amount of purchase information are selected as neighbors. These characteristics of Customer-net make each customer might keep better constituent neighbors and can obtain more current preference related recommendations.
    Experimental results support that Customer-net is a viable solution for ubiquitous personalization services. We showed that the personal purchase amount in achieving efficient and effective recommendation is important when selecting neighbors. Moreover, Customer-net’s recommendation quality improves through the network re-formation process. These results indicate that an active use of less but current information improves recommendation performance in the ubiquitous environment. It is an important result because the number of items and customers grow fast, and mobile devices carried by customers have inherently a limitation of computing power.
    Customer-net also offers remarkably higher quality of recommendations than the CF system and the bestseller system, which results from Customer-net’s accelerated reflection of the dynamics of the customer preference changes. Moreover Customer-net works dramatically faster than the CF system. Customer-net makes it possible for customers to buy products with much less effort in information search and much lower connection time. Customer-net can provide retailers in a physical space with networking platform consisting of customers who can self-manage their information with their personal devices.
  • Index terms
  • Keywords: ubiquitous, recommender system, personalization, local network Keywords: 유비쿼터스, 추천시스템, 개인화서비스, 로컬 네트워크
  • List of digital content of this reports
데이터를 로딩중 입니다.
  • This document, it is necessary to display the original author and you do not have permission
    to use copyrighted material for-profit
  • In addition , it does not allow the change or secondary writings of work
데이터 이용 만족도
자료이용후 의견
입력
트위터 페이스북
NRF Daejeon
(34113) 201, Gajeong-ro, Yuseong-gu, Daejeon, Korea
Tel: 82-42-869-6114 / Fax: 82-42-869-6777
NRF Seoul
(06792) 25, Heonreung-ro, Seocho-gu, Seoul, Korea
Tel: 82-2-3460-5500 / Fax: 82-2-3460-5759
KRM Help Center
Tel : 042-710-4360
E-mail : krmcenter@nrf.re.kr / Fax : 042-861-4380