RPTune is an end‑to‑end framework that improves in‑context catalog search for small merchant businesses by learning to curate product catalogs and fine‑tuning large language models (LLMs) with catalog‑grounded supervision. It uses an encoder‑reorganizer curator to order and prune products based on LLM feedback, and then applies context‑relative rewards during LLM post‑training. Across seven real merchants and 100 complex conversational queries per merchant, RPTune boosts search accuracy by up to 31.4 percentage points from curation alone and an additional 10.3 points on average from post‑training.
By Chuxuan Hu, Hejie Cui, Norman Huang, Shubham Kumar Bharti, Wang-Chiew Tan, Sercan \"O. Ar{\i}k
arXiv:2209. 04942v2 Announce Type: replace-cross Abstract: Problem definition: This paper studies the problem of estimating consumer preferences from bundle sales data.
By Ningyuan Chen, Setareh Farajollahzadeh, Qingwei Jin, Fanni Shen, Guan Wang
The paper proposes a shrinkage estimator that blends store‑specific demand estimates with pooled information from other stores under a multinomial logit model, addressing the trade‑off between local data scarcity and cross‑store heterogeneity. By leveraging self‑concordant‑like properties of the MNL likelihood, the authors derive finite‑sample performance guarantees and regret bounds for assortment and pricing decisions. Numerical experiments and an empirical study on European apparel retailer data show that this transfer‑learning approach improves out‑of‑sample product prioritization and revenue capture compared to both local estimation and full pooling.
By Recep Yusuf Bekci, Stefanus Jasin, Joline Uichanco
arXiv:2507. 10834v4 Announce Type: replace Abstract: Assortment optimization seeks to select a subset of substitutable products, subject to constraints, to maximize expected revenue.
By Guokai Li, Pin Gao, Stefanus Jasin, Zizhuo Wang
The paper investigates how AI shopping assistants influence retail assortment planning by mediating product visibility. It models two mediation types—deterministic ranking and stochastic Plackett‑Luce ranking—and shows that ignoring AI mediation can lead to significant revenue losses (2.52%–11.37%). A simpler deterministic plug‑in approach reduces losses to at most 0.53%, and the authors prove that deterministic mediation allows a fully polynomial‑time approximation scheme for assortment optimization when customer segments are fixed.
By Omar El Housni, Marouane Ibn Brahim
The paper introduces a robust framework for assortment optimization that addresses distributional shifts in customer choice behavior. It demonstrates computational tractability when the nominal choice model is known and develops statistically optimal algorithms for the data‑driven setting, providing matching upper and lower bounds on sample complexity. The authors identify "robust item‑wise coverage" as the minimal data requirement for efficient robust learning, bridging robustness and statistical efficiency in assortment planning.
By Miao Lu, Yuxuan Han, Han Zhong, Zhengyuan Zhou, Jose Blanchet