Extracting Characteristics of Items Based on Patterns in Recommendation Graphs
Daisuke Kitayama and Kazutoshi Sumiya
International MultiConference of Engineers and Computer Scientists (IMECS 2014), pp.332-336, Hong Kong, 2014年3月, 査読有
On online shopping sites such as Amazon and Rakuten, recommended items are displayed along with the items being viewed. We consider that certain recommended items reflect the characteristics of the viewed item. For example, "DVD-R" may be recommended with "Printer," whereas "Printer" might not have the recommended item "DVD-R." In this case, we may assume that the item "Printer" can print a label on a "DVD-R." Thus, a set of items can be expressed as a directed graph structure, which comprises items as nodes and recommendation relations as edges. In this paper, we propose a method for extracting item characteristics based on the patterns in recommendation graphs.We also present an evaluation based on comparisons among items.