BPrice: An Optimal Price Recommendation Framework Based on Product Identification

Abstract

The price of a single product can vary significantly from one retail store to another in the Hong Kong market. The gap between the wealthy and the poor in Hong Kong continues to be large, leading to much of people having to look for bargains daily. While one may be able to search the web for the lowest price, the information on the web is scattered, and it is necessary to compare product names word-for-word. This process is exceptionally inconvenient for some older adults and people with disabilities. The Hong Kong society is pressing for an application that can tackle this problem. In this paper, we propose a framework to recommend the optimal price of the product in the Hong Kong market. The framework aims to speed up the price search with high accuracy. The problem is decomposed into name recognition and database query. The former is accomplished by utilizing the VGG-16 Convolutional Neural Network model pre-trained on the ImageNet dataset via transfer learning and feature matching with Scale-Invariant Features Transform (SIFT) and the colour histogram. The latter is performed on a daily updated MySQL database. The results of the recommendations were satisfactory in the experiments, with the accuracy of validating the CNN model reaching 90.614%.