A restaurant recommendation method that combines neural network algorithms and information extraction from electronic word of mouth

Abstract

Recommendation systems are popular information systems that help consumers manage the information overload problem, encountered when making decisions with many alternatives. One type of such systems are restaurant recommenders. Whilst personality has been recognized as important for influencing consumers’ choice, it has not yet been exploited in restaurant recommendation with dominant recommendation technique in this sector being collaborative filtering. This work proposes a restaurant recommendation method that exploits personalities, opinions and preferences of customers that are either extracted directly from online review platforms or derived from text in electronic word of mouth (eWOM), using information extraction techniques. The proposed method introduces the concept of venue personality based on theory stipulating that customers are more satisfied with brands that have similar personality to their own. A novel deep learning model is designed to extract personality from text and is used to assess customers and venues’ personalities. Food preferences are extracted from eWOM via a custom named-entity recognizer. Customer opinions are extracted from eWOM’s text through topic modelling. Latent information regarding customer preference are extracted form the interactions of customers with restaurants via a neural collaborative filtering (NCF) component. The embeddings of the NCF component are extracted and combined with the above features to form the input based on which an extreme gradient boosting (XGBoost) model is trained to predict restaurants ratings. Results are compared to baseline models using various performance metrics and verify that the method improve restaurant recommendation.