Context-aware neural collaborative filtering

Abstract

Collaborative filtering is a method widely applied by the recommendation system. The basic idea of collaborative filtering is prediction using similar tastes of customers. Mostly, the collaborative filtering method is implemented in two dimensions, [user * item]. Because of using only two dimensions, there is a limitation in Hit ratio. In order to raise Hit Ratio of Collaborative filtering, adding context information (e.g., time, place, people, etc.) is one of the solutions. In this paper we contribute these things. For adding context information we modified the original deep learning program which was published in neural collaborative filtering thesis. After adding context information, Hit ratio is raised from 38% (2 dimension [user * item]) to 75%. We proposed how to estimate the hit ratio of additional context dimensions. Using our estimation, marketers can calculate the hit ratio before running all the combinations of context information. In traditional marketing, marketers should do Segmentation, targeting and finding proper products with tons of data. But our method perform the similar tasks, so marketers can save time and energy