Selective questions influence shopping assistant recommendations and utility
Selective Elicitation as a Commercial Influence Channel: A Reproducible Synthetic Shopping-Agent Stress Test
Machine LearningComputers and Society
Summary
Shopping assistants often ask users questions to learn their preferences before recommending products. This paper shows that by changing which questions the assistant asks, a sponsor can influence the recommended product without altering the final ranking algorithm. The authors tested this idea in a simple simulated setup and found that targeted questions can increase the chance of recommending a favored item but may reduce overall user satisfaction. The study highlights a subtle way of influencing recommendations, but it does not claim to show how common this is in real-world shopping.
What this means in practice
- •For ecommerce platform designers: Assess the impact of question selection on product recommendations under commercial influence to improve transparency.
- •For online advertising teams: Evaluate how targeted preference questions affect sponsored product recommendation likelihood in controlled simulations.
Tested on simulated data.
Authors
Jiapeng Li
Abstract
A commercial incentive need not enter the final ranking algorithm to affect a shopping assistant's recommendation: it may instead influence which preference question the assistant asks. We make this distinction experimentally observable in a deliberately small, synthetic setting. Each task has two products, three verified numerical attributes, a price limit, and a private fixed preference vector. An honest simulated user answers one pairwise question. A separate recommender receives the products and this answer but not the sponsorship assignment. We contrast a neutral question, a soft commercial instruction, and an explicitly adversarial instruction to ask about the sponsor's advantage while omitting the rival's advantage. Across 40 held-out sponsorship-assignment cases (20 distinct catalog-preference contexts), the soft instruction changes no selections. The targeted instruction raises sponsored selection by 0.30 and reduces mean synthetic utility by 0.0547 relative to neutral questioning (95% context-bootstrap interval [-0.0828, -0.0291]) for one language-model recommender. A fixed Bayesian recommender shows a similar effect; a second model makes the same choices on all 120 frozen question-answer inputs. A terminal-answer consistency judge rates all 20 sampled targeted answers consistent, although five have synthetic regret above 0.05; a separate question-coverage dimension flags their one-sided elicitation. A robust partial-preference certificate remains valid under the stipulated synthetic utility but certifies only 16 of 40 targeted cases and is not better than asking a neutral question directly. These results establish neither typical behavior under advertising incentives nor effects on actual consumers.