Fuzzy formal concept analysis based opinion mining for CRM in financial services

Display Omitted Presents a novel complaints analytics hybrid model based on fuzzy formal concept analysis and sentiment analysis.Performs aspect- and concept-level sentiment analysis on complaints.Solves two case-studies on two important domains viz. banking and insurance using real-life datasets.Proposes a novel interactive visualization approach to explore complaints and their main constituents.Partial evaluation by human annotators yielded 64.06% matching score in terms of the opinions determination of aspects. Owing to the easy access to social media, consumers or customers are increasingly turning to social media to express their grievances and feedback on various products and services offered by the Banking, Financial, Services and Insurance industry. Because non-redressal of complaints eventually leads to customer churn, there is an urgent need to analyze the complaints. In this regard, we propose a novel descriptive analytics model that performs complaints/grievances analytics and summarizes the lengthy and verbose complaints concisely in a form that resembles association rules. The proposed hybrid model comprises fuzzy formal concept analysis and concept-level sentiment analysis (FFCA+SA) in tandem, which in turn is compared against formal concept analysis and concept-level sentiment analysis (FCA+SA). Because of the immediate fallout of the negative sentiments, a financial company is interested in studying them in more detail than the positive ones. Therefore, the model generates a list of association rules, the corresponding negative sentiment score along with the list of associated documents. Association rules are rank ordered according to the negative sentiment score, which in turn reflects severity affected services/products. The proposed model also provides interactive visualization that enables business analysts and managers to access a specific set of complaints without having to go through the entire set thoroughly. This saves a lot of time that would have otherwise been spent on cumbersome manual operations. Moreover, partial evaluation of the proposed methodology by human annotators yielded 64.06% matching score in terms of the opinions determination of aspects.