Incorporating Consumer Browse Data: Extended Item-Based Top-K Recommendation Algorithms

Abstract

Personalized recommendation systems are becoming increasingly popular in e-commerce. One of the core recommendation algorithms is item-based recommendation algorithm, which recommend by finding products that are similar to other products using consumers' purchase data. However, purchase data contains too little information to reveal the real preference of consumers. The purpose of this paper is to extend the item-based top-k recommendation algorithm by incorporating consumers' browse data, which contains more useful information in revealing the preference of consumers. First, we extend the cosine-based method of modeling the similarity matrix between two products and construct an extended model subjected to varying conditions in terms of browse and purchase combination. Then we test our extended algorithm on a simulated data set and apply it to a real world data set provided by an online B2C company. Both the results of the simulation data and the real data show that our extended algorithm performs significantly better than the traditional item-based algorithm, in terms of hit rate (HR) and average reciprocal hit rank (ARHR). Specifically, our extended algorithm generates 113, 089 (4349.58%) more sales than random algorithms and 25, 335 (28.86%) more sales than traditional cosine-based algorithms. We discuss the theoretical and managerial implications in the conclusion part.