CRAMER: Control via Request-Aware Masking for Editing Recommenders

Zhiyuan Su, Naihe Feng, Luther Qin, Ga Wu

TL;DR: We propose CRAMER, a request-aware masking framework that adapts frozen sequential recommenders to natural-language requests in real time, achieving controllability and efficiency without retraining.

Abstract: Sequential recommendation models, while powerful, have limited flexibility in responding to immediate user requests, making it difficult to adapt their recommendations to the user's timely interests. Unfortunately, existing user request adaptation methods often incur high computational overhead due to either 1) retraining the entire backbone network or 2) leveraging the inference ability of large language models (a.k.a. prompt engineering), limiting their applicability in large-scale recommendation services. This paper presents **C**ontrol via **R**equest-**A**ware **M**asking for **E**diting **R**ecommenders (**CRAMER**), a framework that takes users' natural-language requests to immediately change sequential recommendation models' behavior. Specifically, inspired by the model control theory, CRAMER treats user requests as control signals to modulate frozen backbone parameters through masking, achieving instant adaptation to diverse requests while avoiding costly retraining. Experiments on multiple large-scale benchmark datasets show that CRAMER outperforms four state-of-the-art request-aware baselines across multiple recommendation metrics while achieving minimal overhead. Moreover, the proposed framework exhibits enhanced controllability and cross-domain adaptability, establishing a new paradigm for request-aware sequential recommendation.

Lay Summary: Online recommendation systems usually learn from what a person has done in the past, but they can be slow to react when the person suddenly asks for something different, such as “show me more exciting games” or “avoid romantic movies.” This paper introduces CRAMER, a way to let an existing recommendation system respond to such requests immediately. Instead of retraining the whole system or asking a large language model to reason through every list of choices, CRAMER uses the request to temporarily adjust which parts of the existing system are used for the next recommendation. This adjustment is lightweight and disappears after the request, so the original system is not permanently changed. Across several large recommendation benchmarks, CRAMER improved the quality of recommendations compared with recent request-aware methods while adding little extra computing cost. The work suggests a practical path toward recommendation systems that are both efficient and more responsive to what users want right now.

Link To Code: https://github.com/zhiyuansu0326/CRAMER-ICML2026

Primary Area: Applications

Keywords: Controllable Model Editing, Request-Aware Masking, Parameter-Efficient Adaptation, Sequential Recommendation

Submission Number: 16252