Learning Custom Experience Ontologies via Embedding-based Feedback Loops

要旨

Companies and organizations rely on behavioral analytics tools like Google Analytics to monitor their digital experiences. Making sense of the data these tools capture, however, requires manual event tagging and filtering---often a tedious process. Prior research approaches have trained machine learning models to automatically tag interaction data, but they draw from fixed digital experience vocabularies which cannot be easily augmented or customized. This paper introduces a novel machine learning feedback loop that generates customized tag predictions for organizations. The approach uses a general experience vocabulary to bootstrap initial tag predictions on interactive Sankey diagrams representing user navigation paths on a digital asset. By interacting with the path visualization, organizations can manually revise predictions. The system leverages this feedback to refine an organization's experience ontology, computing custom word embeddings for each of its terms via vector space refinement algorithms. The updates made to the custom experience ontology and its associated word embeddings result in better event tag predictions for that organization in the future. We conducted a needfinding interview with web analytics professionals to ground our design choices, and present a real-world deployment that demonstrates how, even with just a few training examples, custom tags can be predicted over new data.

著者
Ali Zaidi
Inc., San Francisco, California, United States
Kelsey Turbeville
UserTesting, Inc., San Francisco, California, United States
Kristijan Ivančić
Inc., San Francsico, California, United States
Jason Moss
UserTesting, Inc., Atlanta, Georgia, United States
Jenny Gutierrez Villalobos
UserTesting, Inc, San Franciscoe, California, United States
Aravind Sagar
User Testing, Inc., San Francisco, California, United States
Huiying Li
UserTesting, Inc., San Francisco, California, United States
Charu Mehra
UserTesting, Inc., San Francisco, California, United States
Sixuan Li
UserTesting, San Francisco, California, United States
Scott Hutchins
UserTesting, Inc., San Francisco, California, United States
Ranjitha Kumar
University of Illinois at Urbana-Champaign, Urbana, Illinois, United States
論文URL

https://doi.org/10.1145/3586183.3606715

動画

会議: UIST 2023

ACM Symposium on User Interface Software and Technology

セッション: Interface Evolution: Learning, Adaptation, Customisation

Gold Room
7 件の発表
2023-11-01 23:10:00
2023-11-02 00:50:00