Paper Number

ICIS2025-1536

Paper Type

Complete

Abstract

The increasing availability of digital trace data offers opportunities for representational richness in modeling user behavioral patterns with downstream implications in socio- technical contexts, but poses challenges due to its multidimensionality, longitudinality, and sparsity. We propose a Bayesian tensor-based framework, including a novel empir- ical Bayesian tensor decomposition method, that represents the interplay between users, platform/sensor-based channels, and time using a three-dimensional tensor. Collectively, our framework and method capture user behaviors via low-dimensional latent factors using a family of sparsity-induced priors, thereby enhancing explanatory and predic- tive power. Simulations show our algorithm outperforms alternatives with less bias and stronger predictive power. We further demonstrate practical value through two large- scale case studies—online customer journeys from omnichannel clickstreams and patient behavior in telehealth—and an expert study with healthcare professionals. Together, these findings highlight the framework’s ability to produce relevant, interpretable, and action- able insights, underscoring its broader implications for computational design in socio- technical environments.

Comments

25-Research

Share

COinS
 
Dec 14th, 12:00 AM

Empirical Bayes Tensor Decomposition: A Holistic and Interpretable Representation of Digital Trace Patterns

The increasing availability of digital trace data offers opportunities for representational richness in modeling user behavioral patterns with downstream implications in socio- technical contexts, but poses challenges due to its multidimensionality, longitudinality, and sparsity. We propose a Bayesian tensor-based framework, including a novel empir- ical Bayesian tensor decomposition method, that represents the interplay between users, platform/sensor-based channels, and time using a three-dimensional tensor. Collectively, our framework and method capture user behaviors via low-dimensional latent factors using a family of sparsity-induced priors, thereby enhancing explanatory and predic- tive power. Simulations show our algorithm outperforms alternatives with less bias and stronger predictive power. We further demonstrate practical value through two large- scale case studies—online customer journeys from omnichannel clickstreams and patient behavior in telehealth—and an expert study with healthcare professionals. Together, these findings highlight the framework’s ability to produce relevant, interpretable, and action- able insights, underscoring its broader implications for computational design in socio- technical environments.

When commenting on articles, please be friendly, welcoming, respectful and abide by the AIS eLibrary Discussion Thread Code of Conduct posted here.