The so-called “Anna Karenina Principle” – named for the first like of Tolstoy’s novel (“All happy families are alike; each unhappy family is unhappy in its own way.”) – states that failure can occur in any number of ways, while success tends to be relatively uniform.

This principle has been observed in fields as disparate as evolutionary biology and market research. In this talk, the principle is applied to an NLP task via the intersection of topic and sentiment in patient experience comment data, to both illustrate the trend and propose an approach for its measurement. Consequences of the observation in terms of both imbalances by class for sentiment accuracy and consequent challenges for effective healthcare interventions are discussed.