Beast code in Giters

FED4's repositories

This project attempts to build new product recognition and matching system to replace part of the existing recommendation system. This system can extract all the food products from a web page and match them with current product records in the company database. Furthermore, it can predict the category and nutrition values of the matched product. The decision recommendation model relies on the output nutrition value and category to produce a customized serving suggestion and recommended alternative products based on the personal genetic profile of the user. This system consists of a translator and two BERT based text classifiers. It outperforms the original DLSTM based architecture in accuracy, adaptability and scalability. Customized data sets are created from business-related real-world examples to train and evaluate a BERT based model with enhanced flexibility.

Language:Jupyter Notebook100