Developing Knowledge Graph Interfaces for Intelligence Analysts with the Laboratory for Analytic Sciences
UX Research, User Interviews, UX Design, UI Design
Designed with fellow students Carl Broaddus and Molly Nunes
This project, conducted through the NC State Laboratory for Analytic Sciences, explores the ways in which an intelligent interface could aid intelligence analysts in problem solving through the use of a knowledge graph. The project involved multiple rounds of user interviewing, persona and user journey map development, critiques with users, and a final scenario video demonstrating use of the interface.
Initial research and user interviews
The project began with general research into the work of intelligence analysts to begin to understand their work and the type of information they deal with. Following initial research, we interviewed current and former intelligence analysts to better understand their workflows, typical tools used, daily concerns, and pain points. Analysts from entry-level to senior-level analysts with decades of experience were interviewed over the course of several group and individual sessions. Our team focused on the persona of Tracy, an experienced intelligence analyst with an expertise in languages.
The job of an intelligence analyst is both complex and varied, and can be difficult to picture without observing their work first-hand. However, because of privacy concerns, many specifics about the job could not be shared. We worked to form as clear a picture of their habits, tools, and concerns as we could.
User Journey Map
Combining research and interviews, we created a user journey map and reviewed it with stakeholders to assure we had the best possible picture of their day-to-day work.
Design Phases and User Critiques
The next stage involved three rounds of sketching, wireframing, and critique with expert stakeholders and prospective users. The designs focused on three primary pain points identified in our interviews:
Our sketches and eventual designs focused on how best to address these three pain points while providing an experience that was novel and that leveraged the benefits of a knowledge graph.
All designs pulled from real information from a real-world case file. It was critical throughout the design process to focus on precisely what the user would want to do or know at each step of the research process, and how the interface could best enable that activity. To that end, we also routinely discussed and critiqued designs with our expert stakeholders.
Tech Expert 1:1
As our design centers around the technology of the knowledge graph and implements machine learning to provide insights to users, we wanted to make sure our designs were technically feasible. We consulted a machine learning expert with the Laboratory for Analytic Sciences to discuss our designs and ensure that the machine learning-based strategies employed were realistic. In a real implementation setting, this process should be a continual dialogue between designers and developers; however, for the purposes of this speculative project, a consultation sufficed.
Final Design Overview
The ultimate design resembles a knowledge graph, but adds intelligent filter and sort features that can allow the user to continually refine information to delve into in greater detail.
The user can sort data points by time, relation, and other variables. The smart interface reconstructs the knowledge graph visualization to adapt to analysts' immediate needs. An option to re-sort the knowledge graph by locking in entities also allows for more focused investigations within the workspace. Paths of interactions through which analysts have discovered information useful to solving an investigation are saved and analyzed to provide improved machine-learning-based suggestions in the future.
Filter and Sort for Improved Search
A key element of the design is the filter and sort panel, which provides ways in which the user can reduce the amount of information visible to improve search functionality. The design assumes that the user may or may not know exactly what they are looking for, but may know that they are interested in a certain time window, data type, or for information related to something they have already saved.
The Complete Interface
Screenshots of the complete interface show the graph sorted by time, with the user making use of relational sort features and sliders which limit the timeline based on the length of an event of interest.