Resonance analysis is an evolving text mining method for estimating how much affinity exists in a broader population for a specific group. The method involves a multistage procedure for scoring words according to how distinctive they are of authors in the specific group, and then scoring a broader population according to whether they make similar distinctive choices. While likely applicable to many different forms of written content, resonance analysis was developed for short-form social media posts and has been tested primarily on Twitter and Twitter-like data.
In this working paper, we describe resonance analysis and provide detailed guidance for using it effectively. We then conduct an empirical test of real-world Twitter data to demonstrate and validate the method using tweets from Republican Party members, Democratic Party members, and members of the news media. The results show that the method is able to distinguish Republicans' tweets from Democrats' tweets with 92-percent accuracy. We then demonstrate and validate the method using simulated artificial language to create controlled experimental conditions. The method is very accurate when minimum data requirements have been met. In an operational field test, resonance analysis generated results that mirror real-world conditions and achieved statistically significant agreement with double-blind human analyst judgments of the same users. Finally, we provide examples of resonance analysis usage with social media data and identify opportunities for future research.
Table of Contents
Context and Conceptual Framework
Empirical Validation Test Results
Simulation-Based Validation Test Results
Resonance Analysis in Practice: Closing Thoughts