Home » Research Overview » Understanding Users

Understanding Users

This page highlights a selection of my research work in understanding and learning from users, where I explore how machine learning techniques can be used to study user interactions and infer task intent from patterns. As users engage with interactive systems — whether through clicks, gestures, or natural language — they leave behind rich behavioral signals. By modeling these patterns, we can move beyond surface-level interactions to uncover the underlying goals that drive user behavior. This enables the development of adaptive systems that anticipate needs, streamline workflows, and ultimately create more intuitive, human-centered experiences.

Castor, Pollux, and Gemini

Castor, Pollux, and Gemini are three different implementations of a computational pipeline that projects high-dimensional data into an interactive visualization, combining both dimension reduction and clustering algorithms in the projection. Semantic interactions are implemented within the interface, allowing an analyst to provide implicit task feedback to an underlying machine learning component. With this feedback, an analyst can interact directly with the data to express their interests, incrementally training those underlying models and causing future updates to the visualization to better reflect the dimensions of interest to the analyst. The overarching goal of this research is to explore both the design and interaction space at the intersection of dimension reduction algorithms, clustering algorithms, and semantic interaction.

Related publications:

StarSPIRE and Semantic Interaction Foraging

StarSPIRE is a visual analytics tool designed to explore collections of documents, leveraging users’ semantic interactions to steer (1) a synthesis model that aids in document layout and (2) a foraging model to automatically retrieve new relevant information. In contrast to traditional keyword search foraging, “semantic interaction foraging” occurs as a result of the user’s synthesis actions. To quantify the value of semantic interaction foraging, we used StarSPIRE to evaluate its utility for an intelligence analysis sensemaking task. We found that semantic interaction foraging accounted for 26% of useful documents found, and it also resulted in increased synthesis interactions and improved sensemaking task performance by users in comparison to only using keyword search. Completing text analysis tasks is a continuous sensemaking loop of foraging for information and incrementally synthesizing it into hypotheses.

Related publication:

Interaction Provenance Research

There is fast-growing literature on provenance research, covering aspects such as its theoretical framework, use cases, and techniques for capturing, visualizing, and analyzing provenance data. As a result, there is an increasing need to identify and taxonomize the existing scholarship. Such an organization of the research landscape will provide a complete picture of the current state of inquiry and identify knowledge gaps or possible avenues for further investigation. My research in this area includes a state-of-the-art (STAR) report, producing a comprehensive survey of work in the data visualization and visual analytics field that focus on the analysis of user interaction and provenance data. The survey and papers discussed can be explored online interactively. My work also proposes a task abstraction framework that consists of a three stage approach, composed of (1) initializing a provenance task hierarchy, (2) parsing the provenance hierarchy by using an abstraction mapping mechanism, and (3) leveraging the task hierarchy in an analytical tool.

Related publications:

SIRIUS, Cosmos, and Andromeda

In addition to Castor, Pollux, and Gemini above, my research has also supported similar tools that vary the user interface and types of data that are supported by the systems. SIRIUS presents a dual-view display, permitting semantic interactions to be applied to both the data objects and their attributes. Cosmos introduces an Elasticsearch component to support larger data sets, modeling the synthesis loop as an interactive spatial projection and the foraging loop as an interactive relevance ranking combined with topic modeling.

Related publications:

Google Scholar icon LinkedIn icon orcID icon