개인화는 개인적인 기호를 바탕으로 각 사용자에게 맞춤화된 컨텐츠를 제공하는 것을 목표로 한다. 이러한 관점에서, 개인화의 핵심적인 부분은 각 사용자의 기호에 적합한 컨텐츠나 상품을 추천할 수 있는 추천기술이라 할 수 있다. 선행연구들은 추천시스템의 중요성을 인지하고 새로운 추천기술을 제안하여 왔다. 여러 추천기술들 중에서 협업필터링은 실무에서 활발하게 연구되고 활용되어 왔다. 그러나, 협업필터링은 종종 희박성 또는 확장성 문제를 겪게 된다. 선행연구들 역시 이 두 가지 문제점의 중요성을 인지하고 그에 대한 여러 가지 해결방안들을 제안하였다. 하지만, 여러 선행연구들은 기존의 사용자-상품 매트릭스 외에 다른 원천들로부터 생성된 추가적인 정보를 이용함으로써 문제점들을 해결하려 함으로 인하여 추가적인 시간과 비용을 요하는 다른 문제를 야기하였다. 본 연구에서는 희박성 문제를 완화하고 추천시스템의 성능을 개선하기 위하여 협업필터링을 위한 새로운 내재적 평가방법을 제안한다. 즉, 본 연구에서는 기존 사용자-상품 매트릭스를 이용하여 사용자 간의 신뢰수준을 측정할 수 있는 내재적 평가법에 기반한 사용자-상품 매트릭스의 보완을 통해 희박성 문제를 완화할 수 있는 방안을 제안한다. 또한, 본 연구에서는 제안하는 방안의 유용성을 평가하기 위한 탐색적 실험 결과를 제공한다.
개인화는 개인적인 기호를 바탕으로 각 사용자에게 맞춤화된 컨텐츠를 제공하는 것을 목표로 한다. 이러한 관점에서, 개인화의 핵심적인 부분은 각 사용자의 기호에 적합한 컨텐츠나 상품을 추천할 수 있는 추천기술이라 할 수 있다. 선행연구들은 추천시스템의 중요성을 인지하고 새로운 추천기술을 제안하여 왔다. 여러 추천기술들 중에서 협업필터링은 실무에서 활발하게 연구되고 활용되어 왔다. 그러나, 협업필터링은 종종 희박성 또는 확장성 문제를 겪게 된다. 선행연구들 역시 이 두 가지 문제점의 중요성을 인지하고 그에 대한 여러 가지 해결방안들을 제안하였다. 하지만, 여러 선행연구들은 기존의 사용자-상품 매트릭스 외에 다른 원천들로부터 생성된 추가적인 정보를 이용함으로써 문제점들을 해결하려 함으로 인하여 추가적인 시간과 비용을 요하는 다른 문제를 야기하였다. 본 연구에서는 희박성 문제를 완화하고 추천시스템의 성능을 개선하기 위하여 협업필터링을 위한 새로운 내재적 평가방법을 제안한다. 즉, 본 연구에서는 기존 사용자-상품 매트릭스를 이용하여 사용자 간의 신뢰수준을 측정할 수 있는 내재적 평가법에 기반한 사용자-상품 매트릭스의 보완을 통해 희박성 문제를 완화할 수 있는 방안을 제안한다. 또한, 본 연구에서는 제안하는 방안의 유용성을 평가하기 위한 탐색적 실험 결과를 제공한다.
Personalization aims to provide customized contents to each user by using the user's personal preferences. In this sense, the core parts of personalization are regarded as recommendation technologies, which can recommend the proper contents or products to each user according to his/her preference. P...
Personalization aims to provide customized contents to each user by using the user's personal preferences. In this sense, the core parts of personalization are regarded as recommendation technologies, which can recommend the proper contents or products to each user according to his/her preference. Prior studies have proposed novel recommendation technologies because they recognized the importance of recommender systems. Among several recommendation technologies, collaborative filtering (CF) has been actively studied and applied in real-world applications. The CF, however, often suffers sparsity or scalability problems. Prior research also recognized the importance of these two problems and therefore proposed many solutions. Many prior studies, however, suffered from problems, such as requiring additional time and cost for solving the limitations by utilizing additional information from other sources besides the existing user-item matrix. This study proposes a novel implicit rating approach for collaborative filtering in order to mitigate the sparsity problem as well as to enhance the performance of recommender systems. In this study, we propose the methods of reducing the sparsity problem through supplementing the user-item matrix based on the implicit rating approach, which measures the trust level among users via the existing user-item matrix. This study provides the preliminary experimental results for testing the usefulness of the proposed model.
Personalization aims to provide customized contents to each user by using the user's personal preferences. In this sense, the core parts of personalization are regarded as recommendation technologies, which can recommend the proper contents or products to each user according to his/her preference. Prior studies have proposed novel recommendation technologies because they recognized the importance of recommender systems. Among several recommendation technologies, collaborative filtering (CF) has been actively studied and applied in real-world applications. The CF, however, often suffers sparsity or scalability problems. Prior research also recognized the importance of these two problems and therefore proposed many solutions. Many prior studies, however, suffered from problems, such as requiring additional time and cost for solving the limitations by utilizing additional information from other sources besides the existing user-item matrix. This study proposes a novel implicit rating approach for collaborative filtering in order to mitigate the sparsity problem as well as to enhance the performance of recommender systems. In this study, we propose the methods of reducing the sparsity problem through supplementing the user-item matrix based on the implicit rating approach, which measures the trust level among users via the existing user-item matrix. This study provides the preliminary experimental results for testing the usefulness of the proposed model.
* AI 자동 식별 결과로 적합하지 않은 문장이 있을 수 있으니, 이용에 유의하시기 바랍니다.
문제 정의
The purpose of this study is to remedy the sparsity problems in CF using implicit trust-based filtering. As aforementioned, implicit trust-based filtering indirectly estimates the trust score using the existing user-item matrix, whereas explicit trust-based filtering needs additional user rating data.
제안 방법
In this study, we propose a novel implicit rating approach for collaborative filtering in order to mitigate the sparsity problem as well as to enhance the performance of recommender systems. In this study, we propose the methods of reducing the sparsity problem through supplementing the user-item matrix based on the implicit trust score, which measures the trust level among users, by utilizing the existing user-item matrix.
In this study, we propose the methods of reducing the sparsity problem through supplementing the user-item matrix based on the implicit trust score, which measures the trust level among users via the existing user-item matrix.
In this study, we propose a novel implicit rating approach for collaborative filtering in order to mitigate the sparsity problem as well as to enhance the performance of recommender systems. In this study, we propose the methods of reducing the sparsity problem through supplementing the user-item matrix based on the implicit trust score, which measures the trust level among users, by utilizing the existing user-item matrix. The preliminary experimental results demonstrate the usefulness of the proposed model.
In this study, we set the average MAE (mean absolute error) as the criterion for evaluating the performances of the comparative models. The MAE is frequently used in CF literature, and represents the difference between the predicted and actual rating of users (Breese et al.
대상 데이터
As aforementioned, we evaluated only 50 users and 20 items for the proposed model. Although this study attained a better recommendation performance compared to the conventional CF, we might be able to obtain more reliable results if we can collect a sufficient amount of samples.
In order to validate the usefulness of our proposed recommendation algorithm, we adopt ‘empirical validation’ that is based on a real-world dataset. Our experiments are run on a famous open dataset, the MovieLens2 datasets. MovieLens is originally a web-based recommender system for movies, which was developed by the GroupLens Research Project at the University of Minnesota.
데이터처리
A typical CF can estimate similarities between users by using the evaluation scores from the user-item matrix, cosine vector or Pearson’s correlation coefficient.
후속연구
As aforementioned, we evaluated only 50 users and 20 items for the proposed model. Although this study attained a better recommendation performance compared to the conventional CF, we might be able to obtain more reliable results if we can collect a sufficient amount of samples.
참고문헌 (5)
Breese, J., D. Heckerman and C. Kadie, "Empirical analysis of predictive algorithms for collaborative filtering," Proceedings of the 14th Conference on Uncertainty in Artificial Intelligence, (1998), 43-52.
Goldberg, K., T. Roeder, D. Gupta and C. Perkins, "Eigentaste: A constant time collaborative filtering algorithm," Information Retrieval Journal, Vol.4, No.2(2001), 133-151.
Papagelis, M., D. Plexousakis and T. Kutsuras, "Alleviating the sparsity problem of collaborative filtering using trust inferences," Proceeding iTrust'05 Proceedings of the Third international conference on Trust Management, (2005), 224-239.
Sarwar, B. M., J. A. Konstan, A. Borchers, J. Herlocker, B. Miller and J. Riedl, "Using filtering agents to improve prediction quality in the GroupLens research collaborative filtering system," Proceedings of the 1998 ACM Conference on Computer Supported Cooperative Work, (1998), 345-354.
Shambour, Q. and J. Lu, "A hybrid trust-enhanced collaborative filtering recommendation approach for personalized government-to-business e-services," International Journal of Intelligent Systems, Vol.26, No.9(2011), 814-843.
AI-Helper
※ AI-Helper는 부적절한 답변을 할 수 있습니다.