Text Based Restaurant Recommendation System using End-To-End Memory Network

With growing use of online content streaming websites, online shopping, and other exclusively online services, it becomes more and more imperative for technology companies to invest a lot of funds into a system to gauge user needs and requirements. To bridge this gap, there has been an influx of recommendation systems in the markets. From advertisements, to movies, and products we buy, recommendation engines are feeding on new data everyday to learn user trends. This paper tries to focus on improving the text based recommendation systems that can be implemented to leverage the vast review data that can be found on websites. We suggest using a novel memory based end-to-end network mechanism to reduce the need for long term dependencies and to reduce the need for memory intensive systems. As we generate more and more reviews and textual data on the web everyday, we need to be able to use this data to make meaningful analytical and business predictions. With the ability to perform multiple lookups, implement attention mechanism and back-propogation, this system was found to perform much better when compared to CNN, RNN and LSTM alternatives in our testing.