Facilitating Knowledge Sharing from Domain Experts to Data Scientists for Building NLP Models

Abstract

Data scientists face a steep learning curve in understanding a new domain for which they want to build machine learning (ML) models. While input from domain experts could offer valuable help, such input is often limited, expensive, and generally not in a form readily consumable by a model development pipeline. In this paper, we propose Ziva, a framework to guide domain experts in sharing essential domain knowledge to data scientists for building NLP models. With Ziva, experts are able to distill and share their domain knowledge using domain concept extractors and five types of label justification over a representative data sample. The design of Ziva is informed by preliminary interviews with data scientists, in order to understand current practices of domain knowledge acquisition process for ML development projects. To assess our design, we run a mix-method case-study to evaluate how Ziva can facilitate interaction between domain experts and data scientists. Our results highlight that (1) domain experts are able to use Ziva to provide rich domain knowledge, while maintaining low mental load and stress levels; and (2) data scientists find Ziva's output helpful for learning essential information about the domain, offering scalability of information, and lowering the burden on domain experts to share knowledge. We conclude this work by experimenting with building NLP models using the Ziva output for our case study. CCS CONCEPTS• Human-centered computing → Empirical studies in HCI ; Interactive systems and tools.

“It’s Like the Value System in the Loop”: Domain Experts’ Values Expectations for NLP Automation

“…The signifcance of subject matter expertise (SME) in the decision making was inline with our research fndings across diferent domains [104,136]. For example, Houde et al [72], who created dystopian Generative AI futures, and Diakopoulos and Johnson [49] who studied ethical implications to Deepfakes emphasised the importance of a human expert (e.g., moderator) to reduce (e.g., through regulation) the harmful impact of these advanced automated technologies.…”

“…Examples include natural language generation of news articles [46], question-answering, query understanding [1], translation and summarization [92], and others. Several of these tools (e.g., AutoML/AutoAI/AutoText) are often developed with dual objectives in mind to facilitate both technical experts and domain experts [35,104,135,137]. For example, with the help of these tools, expert data scientists no longer have to engage in repeated low level programming tasks, while domain experts such as business professionals can build and run ML/NLP models easily [135].…”

Investigating How Practitioners Use Human-AI Guidelines: A Case Study on the People + AI Guidebook

“…Similarly, studies on data science and AI practitioners revealed that they worked to sensitize their cross-functional partners, such as PMs and domain experts, to AI concepts [54,72,78]. The way they worked resembled design practice [54]; they probed their collaborators to get at the underlying problem, to frame and reframe it as a well-defined data science problem [54,72,76]. Finally, research showed that effective AI teams had an approach similar to startups [83], where they would focus on a minimum viable product at the intersection of what is feasible with AI models and data, and what is desirable and viable for customers and users [96,103,105,107].…”

Crystalline: Lowering the Cost for Developers to Collect and Organize Information for Decision Making

“…Sensemaking is widely considered to be the process of searching, collecting, and organizing information to iteratively develop a mental model that best fits the evidence [96,106]. As knowledge workers [9], many activities that developers perform on a daily basis involve extensive sensemaking, such as designing the overall software architecture [56,83], learning and understanding unfamiliar code and concepts [26,73], debugging and fixing incorrect software behaviors [25,74], planning and executing code refactorings [32,41,86], and evaluating past code and design patterns for future reuse [82,91].…”