No abstract

This paper reports on the results of extracting useful information from text notes captured within a Customer Relationship Management (CRM) system to segment and thus target groups of customers likely to respond to cross-selling campaigns. These notes often contain text that is indicative of customer intentions. The results indicate that the notes are meaningful in classifying customers who are likely to respond to purchase multiple communication devices. A Naïve Bayes classifier outperformed a Support Vector Machine classifier for this task. When combined with structured information, the classifier performed only marginally better. Thus, customer service notes can be an important source of predictive data in CRM systems.

As healthcare costs rise, hospitals are seeking ways to improve operations. This paper examines the usefulness of free-form notes to solve a classification problem commonly associated with customer churn. The authors show that classifiers which incorporate free-form notes, using natural language processing techniques, are up to 9% more accurate than classifiers that are solely developed using structured data. The authors suggest that hospitals and chronic disease management clinics can use structured data and free-form notes from electronic health records to predict which patients are likely to cease receiving care from their facilities. Classification tools for predicting patient churn are of interest to hospital administrators; such information can aid in resource planning and facilitate smoother handoffs between care providers.

Predicting patient turnover within health services is beneficial for resource planning. In this chapter, patient turnover is viewed as a form of customer churn. As such, the authors examine whether free-form notes are useful for solving the classification problem typically associated with customer churn. The authors show that classifiers which incorporate free-form notes, using natural language processing techniques, are up to 11% more accurate than classifiers that are solely developed using structured data. In addition, the authors show that free-form notes aggregated for each account perform better than treating each note separately. It is suggested that hospitals and chronic disease management clinics can use structured data and free-form notes from electronic health records to predict which patients are likely to cease receiving care from their facilities. Classification tools for predicting patient churn are of interest to hospital administrators; such information can aid in resource planning and facilitates smoother handoffs between care providers.

Anticipating effects of proposed clinical policies is a difficult task. This study investigates the usefulness of agent-based simulations for evaluating clinical policies. Two policies for continuity of care for patients with type 2 diabetes are investigated using an agentbased simulation. Computational models of a dynamic decision environment were simulated to determine aggregated effects of individual care-providing agents acting to achieve clinical goals. The simulated policies were: 1) continuous care (CC), where each patient was randomly assigned a specific physician model for care across visits; 2) opportunistic care (OC), where each patient on each visit was randomly assigned to a physician model for treatment. These policy scenarios are at the crux of a debate as to whether continuity of care needs to be administered by a single provider or by a single organisation (e.g., clinic). The study determines under which conditions CC and OC policies result in favourable patient outcomes.