Using Traditional Text Analysis and Large Language Models in Service Failure and Recovery
Published in Journal of Service Research
Service failure and recovery (SFR) typically involves one or more people (or machines) talking or writing to each other in a goal-directed conversation.
While SFR represents a prime context for understanding how language reflects and shapes the service experience, this subfield has only begun to apply text analysis methods and language theories to this context.
This tutorial offers a methodological guide for traditional text analysis methods and large language models and suggests some future research paths in SFR. It also provides user-friendly workflow repositories, in Python and KNIME Analytics, that researchers with (and without) coding experience can use. In doing so, the scope is to encourage the next wave of text analysis in SFR research.
Conversation is a central feature of service failure and recovery (SFR). Customers communicate how a product, employee, or firm has failed to meet their needs or expectations. Frontlines (employees and AI) respond in ways they hope will de-escalate or resolve the customer’s complaint.
Given the strategic and economic importance of SFR, understanding how to improve these interactions is a central goal of service research, and a tremendous body of work has helped do so through surveys, experiments, and observational studies. But what about the natural language customers and employees use in service situations?
Analyzing the text produced in in-person, phone, or digital service can tell us a lot about the people and processes in SFR. It can provide novel insight into people’s emotions, motivations, attitudes, and intentions in this context and inform how communication approaches shape customer and employee attitudes and behaviors.
This tutorial aims to help scholars use textual data in SFR research. It offers three contributions. First: a practical guide for using five important text analysis methods in SFR research, including reusable text mining workflows (i.e., code) in Python and KNIME, to help make these methods accessible to nonspecialists (and noncoders). Second: deeper insights into using newer Large Language Models (LLMs), a category of Generative AI, for text analysis. Lastly, some avenues for future SFR research applying both traditional text mining methods and LLMs.