Deep learning enabled dynamic user user product rating based recommendation system

Abstract

Recommendation systems are the utmost need in any E-commerce business. There are various recommendation systems built with machine learning and deep learning algorithms. In abstract, all the recommendation models fall in one of the following categories. Popularity based model, classification based models, models based on the contents, collaborative filtering models, hybrid approaches that combines two or more models and finally models based on association. It has been observed from the literature that most of the studies follow collaborative filter approach or its variants in one or other way. Collaborative filtering models mainly depend on the similarity between the users based on the rating they make. it is a two step process in which the first step concentrates on finding the similarity based on which the second step finds the rating of the unrated products by the intended user. The first part of this work concentrates on experimenting with the existing models; first experimentation with the Zomato dataset is made. The problem is considered as multi class classification problem. Various machine learning algorithms are implemented. Decision Tree, Random Forest, K-Nearest Neighbor, and a neural network model. It has been observed though the accuracy of the Artificial Neural Networks is low it performs better than the other machine learning models. The accuracy obtained is 68%. The second part of the work concentrates on the implementation of the collaborative filtering approach. In general the recommendations to a user are made based on similarity that exists between the intended user and the other users. This similarity can be calculated either based on the similarity between the user profiles or the similarity between the ratings made by the users. First phase of this work concentrates on experimentally analyzing both these model and get a deep insight of these models. With the lessons learned from the insights, second phase of the work concentrates on developing a deep learning model.