Hierarchically structured recommender system for improving NPS

Net Promoter System (NPS) is well known as an evaluation measure of the growth engine of big companies in the business area. The ultimate goal of my research is to build an action rules and meta-actions based recommender system for improving NPS scores of 34 companies (clients) dealing with similar businesses in the US and Canada. With the given original dataset, data preprocessing has been completed to result in better data representation and quality. The recommender system is built on top of a hierarchical clustering dendrogram which is generated by applying agglomerative clustering algorithm to a matrix of distance mixing semantic similarity and geographical distance by assigning weighted factors. To maximally expand the dataset of a single client by merging it with datasets of other clients under certain conditions, Hierarchically Agglomerative Method for Improving NPS (HAMIS) has been developed and applied to all clients respectively.;To extract meta-actions from customers' comments for triggering generated action rules and achieving desirable effect, a new text mining based strategy has been designed to accomplish tasks involving sentiment analysis, text summarization, feature identification and most importantly meta-action generation. Compared to other relevant works, our method has been proven to be more flexible and has achieved more satisfying results. Considering the fact that commonness, differential benefit and applicability exist when executing groups of meta-actions, a method for searching best groups of meta-actions has been designed to boost the performance of applying meta-actions, which also improves the applicability and practicability of our system.