Effective Natural Language Interfaces for Data Visualization Tools

How many Covid cases and deaths are there in my hometown? How much money was invested into renewable energy projects across states in the last 5 years? How large was the biggest investment in solar energy projects in the previous year? These questions and others are of interest to users and can often be answered by data visualization tools (e.g., COVID-19 dashboards) provided by governmental organizations or other institutions. However, while users in organizations or private life with limited expertise with data visualization tools (hereafter referred to as end users) are also interested in these topics, they do not necessarily have knowledge of how to use these data visualization tools effectively to answer these questions. This challenge is highlighted by previous research that provided evidence suggesting that while business analysts and other experts can effectively use these data visualization tools, end users with limited expertise with data visualization tools are still impeded in their interactions.

One approach to tackle this problem is natural language interfaces (NLIs) that provide end users with a more intuitive way of interacting with these data visualization tools. ... mehrEnd users would be enabled to interact with the data visualization tool both by utilizing the graphical user interface (GUI) elements and by just typing or speaking a natural language (NL) input to the data visualization tool. While NLIs for data visualization tools have been regarded as a promising approach to improving the interaction, two design challenges still remain. First, existing NLIs for data visualization tools still target users who are familiar with the technology, such as business analysts. Consequently, the unique design required by end users that address their specific characteristics and that would enable the effective use of data visualization tools by them is not included in existing NLIs for data visualization tools. Second, developers of NLIs for data visualization tools are not able to foresee all NL inputs and tasks that end users want to perform with these NLIs for data visualization tools. Consequently, errors still occur in current NLIs for data visualization tools. End users need to be therefore enabled to continuously improve and personalize the NLI themselves by addressing these errors. However, only limited work exists that focus on enabling end users in teaching NLIs for data visualization tools how to correctly respond to new NL inputs.

This thesis addresses these design challenges and provides insights into the related research questions. Furthermore, this thesis contributes prescriptive knowledge on how to design effective NLIs for data visualization tools. Specifically, this thesis provides insights into how data visualization tools can be extended through NLIs to improve their effective use by end users and how to enable end users to effectively teach NLIs how to respond to new NL inputs.
Furthermore, this thesis provides high-level guidance that developers and providers of data visualization tools can utilize as a blueprint for developing data visualization tools with NLIs for end users and outlines future research opportunities that are of interest in supporting end users to effectively use data visualization tools.