Comparing Filtering Techniques in Restaurant Recommendation System

This paper studies the key analytics of the restaurant recommendation system, namely predicting restaurant satisfaction rating based on customer and restaurant characteristics. As food industry grows and offers more variety of restaurants, customers generally have difficulty discovering a restaurant that suits or satisfies them. This paper aims to predict restaurant satisfaction rating based on three methodologies: content-based filtering, collaborative filtering, and hybrid filtering. For content-based filtering, this paper proposes using regression to create a prediction model from customer and restaurant characteristics. For collaborative filtering, our proposed model employs a combination of cluster analysis, similarity test, and weighted sum in order to analyze factors that influence the satisfaction rating. Cluster analysis helps to reduce the impact of sparsity in collaborative filtering. Subsequently, hybrid filtering is proposed to combine the results from the two techniques above to generate the final rating. Our results have shown that hybrid filtering outperforms content-based filtering using regression model and collaborative filtering using cluster-based technique.