Pricing recommender system for E-Commerce platforms

Abstract

E-commerce has drastically scaled up in recent years. As it keeps growing, the competition gets stiffer, and thus, many of the tasks of an e-commerce system requires to be efficiently automated. Some requirements for an e-commerce business platform are recommending similar products, recommending product tags based on user searches, market-basket, and association analysis, promote customer-targeted discount offer, price comparison of similar product from other sites, etc. One such requirement, which is what we worked on in this project is a price recommender system. A dynamic pricing module should consider as many as possible factors that impact the price of a product, and then suggest a price to the management that is optimal. We try to portray a dynamic product pricing tool, which is a long-time attraction for the researchers to put an end to price rigidity. While other recommendation systems act as an add-on service to customers, price recommender can serve the business owners to strategize pricing decisions in the long run. Optimal pricing by any e-commerce not only enhances revenue in the long run, but also should increase business popularity, and provide a competitive edge to the business owners. Information externalities and a large number of pricing factors is often difficult to deal by a pricing team of experts. If the entire market-statistics is not captured, by mere expertise in today’s data-driven world, the set prices may not always be optimal. So, knowledge-based recommender should be considered as a handy tool or service as the way we have proposed it. It starts with a barebone model, but with increased learning about pricing factors and data, we can build it in a modular way with careful consideration about model and market-related assumptions. In this study, we obtained a sales dataset, performed data preprocessing before fitting some machine learning models, and tried finding price recommender on the test set of data. We present our analysis and findings suggested how these models can be fitted into the context of Bangladesh and what possible improvements can be performed to further enhance the accuracy and efficiency of a price recommender.