Combo-Recommendation Based on Potential Relevance of Items
Combo recommendation expects to recommend a collection of products to users in a Groupon way.The representative application is combo recommendation in the travel industry,which is also called package recommendation and may include different landscapes according to the inherent features.Compared with traditional recommendation scenario,combo recommendation has the following characteristics: (1) sparsity: information for combos is much less than that for individual items;(2) collectivity: every combo is composed of multiple individual products with different features;(3) diversity: products composed of combos may have different features;(4) relevance: products inside combos have some kind of potential relevant.Traditional recommendation algorithms may perform poor for they consider nothing about these four characteristics in the models.Aiming at improving performance of combo recommendation,our work proposes a novel combo recommendation algorithm called RBM-CR based on the Restricted Boltzmann Machine.RBM-CR algorithm takes advantage of usersconsumption histories to derive the correlations among products by mapping from visible features to hidden features,and to profile users and combos by those hidden features.Finally,experiments on real dataset verify effectiveness and accuracy of our algorithm.
Yanhong Pan Yanfei Zhang Rong Zhang
Institute for Data Science and Engineering,Software Engineering Institute,East China Normal University,Shanghai,China
国际会议
International Asia-Pacific Web Conference(第18届国际亚太互联网大会)
苏州
英文
505-517
2016-09-23（万方平台首次上网日期，不代表论文的发表时间）