PersonaBase: Computational Resources for Advancing the Research and Practice of Data-Driven Personas
TL;DR
PersonaBase, a repository of computational DDP resources that gathers tasks, datasets, metrics, and baselines, is presented and it is proposed that future work on DDP resources addresses needs by community building and learning from other computational sciences, while moving closer to full benchmarking solutions.
Abstract
Researchers struggle to replicate studies and measure progress in data-driven persona (DDP) research, while practitioners struggle to understand, develop, and validate DDPs using shared methods and metrics. To address these challenges, we adopt ideas from the natural language processing (NLP) community to present PersonaBase, a repository of computational DDP resources that gathers tasks, datasets, metrics, and baselines. The first version of PersonaBase contains 11 notebooks demonstrating algorithmic persona generation and evaluation approaches, 27 persona generation prompts for large language models (LLMs), 19 systems, and 12 datasets to experiment with. We describe these DDP resources and evaluate target users’ first-impression utility with two formative studies, one with 4 persona researchers and another with 25 industry practitioners. The evaluators generally found the resources useful and relevant, with more experienced DDP users identifying datasets and LLM prompts as particularly relevant. Both researchers and practitioners exhibited needs for (1) improved documentation and onboarding, (2) reproducible pipelines, (3) template-based resources, and (4) improved accessibility for non-technically oriented users. We propose that future work on DDP resources addresses these needs by community building and learning from other computational sciences, while moving closer to full benchmarking solutions.
