Relative Price Recommendation: Implementing Category Relations and Price of Products to Map Preference

Abstract

In an e-commerce setting, a successful recommender system needs to incorporate the needs of consumers based on previous purchases by users. Current recommender systems incorporate different features and images of products to recommend products to consumers. Interestingly, the price of an object is one of the biggest constraints that the user faces before making a purchase. However, there is a research gap in our understanding of how to incorporate price into recommender systems.

This thesis explores the price aspect of products and how to incorporate price into a relative price-based recommendation. This work is different from modern approaches that observe the price elasticity and price sensitivity of products and to understand consumer behavior. This thesis will highlight how price as a comparable feature can be used to understand consumer interests and how relative price can be used to help narrow down products a consumer will be interested in.

This thesis will initially observe the performance of classic models such as user recommender and latent factor models such as Probabilistic Matrix Factorization. Then I will combine the category of relationships based on an economic theory of substitutes and complements to improve the accuracy of currently used models. This framework will address the issues of the long-tail problem inherent in data distribution.

From testing with Amazon review data, it has been observed that my framework can levitate the long-tail problem inherent in the dataset. Combined with previous works on price sensitivity, my framework can be used to explain purchase strategies of consumers along with consumer interest.