A DeepFM model-based personalized Restaurant Recommendation System

Abstract

Due to the continuously growing and overwhelming number of choices on the internet, it is incredibly challenging to extract relevant and meaningful information. So, there is a strong need to efficiently filter, gather, and prioritize the most useful information from the internet automatically that causes to mitigate information overload issues and makes the customers confused. However, the proposed recommendation system solves these issues by filtering a large volume of information and recommending personalized content or products. In this work, a recommendation system technology for food restaurants is applied because there are many choices for the customers to select. Specifically, it becomes challenging for tourists to find a good place to eat. Another significant issue in current systems is that the food is not recommended as per the user's age and gender. To incorporate the factor of age and gender into the recommendation, the proposed system predicts the age and gender of a user before recommending a restaurant. The prediction model predicts gender and age from the user's faces accurately, which supports the process of recommendation to be more accurate and valid. Hence, a DeepFM (Factorization-Machine) based personalized restaurant recommendation mechanism is proposed that recommends restaurants to users according to age and gender more accurately and precisely. Moreover, the technique of web crawling is used to collect restaurant food data, user comments, and ratings from the different websites and trained the DeepFM model. An open-source Zomato restaurant review data set for training Machine Learning (ML) based DeepFM model is used. This DeepFM-based restaurant recommendation system efficiently recommends top-10 restaurants list to users. Experiments were conducted to generate a list of top 10 recommendations based on Zomato open restaurant reviews data, which were found to be more useful.