An Information-Theoretic Law Governing Human Multi-Task Navigation Decisions

TL;DR

This paper uncovers an information-theoretic law governing human multi-task navigation decisions, demonstrating that the likelihood of order inversion correlates with the entropy of item pairings, and proposes a model that accurately predicts shopping sequences.

Contribution

It introduces a novel entropy-based law for human navigation decisions and a noisy distance estimation model that aligns with observed data trends.

Findings

Likelihood of item order inversion is bound by pairwise entropy.

The proposed model reproduces the entropy law with high accuracy.

Model effectively predicts shopping list retrieval sequences.

Abstract

To better understand the process by which humans make navigation decisions when tasked with multiple stopovers, we analyze motion data captured from shoppers in a grocery store. We discover several trends in the data that are consistent with a noisy decision making process for the order of item retrieval, and decompose a shopping trip into a sequence of discrete choices about the next item to retrieve. Our analysis reveals that the likelihood of inverting any two items in the order is monotonically bound to the entropy of the pair-wise ordering task. Based on this analysis, we propose a noisy distance estimation model for predicting the order of item retrieval given a shopping list. We show that our model theoretically reproduces the entropy law seen in the data with high accuracy, and in practice matches the trends in the data when used to simulate the same shopping lists. Our approach has direct applications to improving simulations of human navigation in retail and other settings.