A Recommendation System Based on Multi-attribute

作者：Dong Jiali; Li Xiaoyong; Fang Binxing

来源：9th International Conference on Service Science, ICSS 2016, 2016-10-15 To 2016-10-16.

DOI：10.1109/ICSS.2016.30

• Application environment
• Existing recommendation systems
• Latent factor
• Multi-attributes
• Multi-dimensional space
• Personal recommendations
• Quality of recommendations
• Scientists and engineers

In recent years, more and more internet companies trend to fuse their products with personal recommendation functions. Although the idea of this business model satisfies characteristic of Internet industry felicitously, scientists and engineers gradually find out that the effect is worse than they expected. The quality of recommendation is limited by cold start, data sparse, lack of learning contextual aware, and so on. One of the most important drawbacks of existing recommendation systems is that they usually use only rating matrix as useful information and not fully consider item attributes. In this paper, we proposed an optimized recommendation system, which based on modeling of items in a multidimensional space of item's attribute. The presented approach takes item attributes, user preferences and user context into consideration, and is mainly composed of two modules: (1) attribute-latent factor model, which is used to show user latent preference for each attribute, (2) multi-attribute model, which is used to show user preference between all attributes. The research is conducted in a practical application environment, and the preliminary experimental results show the validity of the recommendation.

• 出版日期2017
• 单位北京大学