Abstract
It is challenging to acquire user data from a diverse set of adequate numbers of suitable real end users for design, evaluation, and improvement of a product. Involving real users in evaluation requires a sufficient level of product maturity. It is difficult to bootstrap and test personalization because it requires personal data. This disclosure describes techniques to overcome these limitations by generating synthetic data that mimics authentic digital and social user practices via simulated personas with local memory. The simulation utilizes the semantic understanding and reasoning capabilities of LLMs and includes interaction between the personas and various digital applications and services. Each persona is associated with a respective device to access various applications and services as well as a simulated virtual assistant. A generic two-step LLM prompting approach can be applied to select a contextually relevant application, followed by the specific application function based on corresponding natural language descriptions. Human narrators can proactively trigger the simulation of specific scenarios by injecting prompts and information delivered to the personas. Persona activity logs can be used for efficient, cost-effective, and flexible product evaluation.
Creative Commons License
This work is licensed under a Creative Commons Attribution 4.0 License.
Recommended Citation
Elyada, Oded; Riva, Oriana; Aharon, Avia; Hakimi, Ido; Ziv, Ron; and Karmon, Danny, "Simulating Authentic User Personas to Generate Analytics for Evaluating Digital Products", Technical Disclosure Commons, (September 23, 2024)
https://www.tdcommons.org/dpubs_series/7372