casa 19(17): e4

Research Article

Item-based recommendation with Shapley value

Download347 downloads
  • @ARTICLE{10.4108/eai.18-3-2019.159341,
        author={Tri  Minh Huynh and Tai  Huu Pham and Vu  The Tran and Hiep  Xuan Huynh},
        title={Item-based recommendation with Shapley value},
        journal={EAI Endorsed Transactions on Context-aware Systems and Applications},
        volume={6},
        number={17},
        publisher={EAI},
        journal_a={CASA},
        year={2019},
        month={6},
        keywords={Collaborative Filtering (CF) Recommender System (RS), Multi-Criteria (MC), Interaction, Decision-Making (DM), importance, Shapley},
        doi={10.4108/eai.18-3-2019.159341}
    }
    
  • Tri Minh Huynh
    Tai Huu Pham
    Vu The Tran
    Hiep Xuan Huynh
    Year: 2019
    Item-based recommendation with Shapley value
    CASA
    EAI
    DOI: 10.4108/eai.18-3-2019.159341
Tri Minh Huynh1,*, Tai Huu Pham2, Vu The Tran3, Hiep Xuan Huynh2
  • 1: Kien Giang University, Viet Nam
  • 2: Can Tho University, Viet Nam
  • 3: University of Science and Technology, Da Nang University, Viet Nam
*Contact email: Hmtri@vnkgu.edu.vn

Abstract

Discovering knowledge in archival data is the goal of researchers. One of them is collaborative filtering recommender system is developing fastly today. It may be rather effective in sparse and "long tail" datasets. Calculating to make decision based on many criteria is really necessary. Relationships, interactions between criteria need to have been fully considered, decision will be more reliable and feasible. In this paper, we propose a new approach that builds a recommender decision-making model based on importance of item, set of items with Shapley value. This model also incorporates traditional techniques and some our new approaches and was tested, evaluated on multirecsys tool we develope from some available tools and uses standardized datasets to experiment. Experimental results show that the proposed model is always satisfactory and reliable. They can be applied in appropriate contexts to minimize limitations of recommender system today and is a research way next time.