What do my users want? Leveraging users insights to improve recommender systems in eWOM communities

eWOM (electronic word-of-mouth) communities not only help their users to gain insights through the exchange of information about products, but also to make the right purchase decisions on the basis of other users' opinions. The vast number of reviews and ratings contain plenty of useful information and recommender systems are an effective tool for filtering them and providing users with the information they are looking for. However, traditional recommender systems use the rating as an input to recommend items, which leads to the cold-start problem and data sparsity. The aim of this paper is to reduce the undesirable outcomes caused by these problems and to optimize the predictive outcomes of the recommendations in the context of eWOM communities. To this end, we propose a hybrid recommender system that combines Social and eWOM variables as an input and uses the Kmeans algorithm for dimensionality reduction and the collaborative filtering SVD++ algorithm to optimize the accuracy of recommendations. Our results show that recommender systems based on users' behavioral data from eWOM communities improve recommendations compared to other recommender systems that use different variables as an input and PCA as a dimensionality reduction technique.