This dissertation presents a novel data-driven approach to solve the problem of improving customer loyalty and customer retention. The data mining concepts of action rules and meta actions are used to extract actionable knowledge from customer survey data and build a knowledge-based recommender system (CLIRS - Customer Loyalty Improvement Recommender System). Also, a novel approach to extract meta-actions from the text is presented. So far, the use of meta-actions required a pre-dened knowledge of the domain (e.i. medicine). In this research an automatic extraction of meta actions is proposed and an implemented by applying Natural Language Pro-cessing and Sentiment Analysis techniques on the customer reviews. The system's recommendations were optimized by means of implemented mechanism of triggering optimal sets of action rules. The optimality of recommendations was dened as maximal Net Promoter Score impact given minimal changes in the company's service. Also, data visualization techniques are proposed and implemented to improve understanding of the multidimensional data analysis, data mining results and interactingwith the recommender systems results. Another important contribution of this research lies in proposing a strategy for build-ing a new set of action rules from text data based on sentiment analysis and folksonomy. This new approach proposes a strategy for building recommendations directly from action rules, without triggering them by meta actions. The coverage and accuracy of the opinion mining was significantly improved within a series of experiments, which resulted in better recommendations. Therefore, the research presents a novel approach to build a knowledge-based recommender system whenever only text data is available.
