AdaptFuse: Training-Free Sequential Preference Inference via Externalized Bayesian Computation

arXiv:2604.03925v2 Announce Type: replace Abstract: Large language models (LLMs) struggle to accumulate evidence across multiple rounds of user interactions, failing to update their beliefs in a manner consistent with Bayesian inference. Existing solutions require fine-tuning on sensitive user interaction data, limiting their applicability in privacy-conscious settings. We propose extsc{AdaptFuse}, a training-free framework that externalizes probabilistic computation entirely from the LLM, where a symbolic module maintains a Bayesian posterior over a discrete hypothesis set, while a frozen LLM contributes semantic reasoning via multi-sample Dirichlet aggregation. The two signals are combined through entropy-adaptive fusion, which automatically weights each source by its predictive confidence, shifting reliance from the LLM to the symbolic posterior as evidence accumulates. We evaluate across three domains: flight recommendation, hotel recommendation, and web shopping; on Gemma2 9B, Llama3 8B, and Qwen~2.5 7B. extsc{AdaptFuse} consistently outperforms both prompting baselines and fine-tuned Bayesian Teaching models on all tasks, with accuracy improving monotonically over interaction rounds. These results demonstrate that principled inference-time algorithms can substitute for fine-tuning in personalized recommendation, without storing or training on sensitive user data.

Related

• BenchPreS: A Benchmark for Context-Aware Personalized Preference Selectivity of Persistent-Memory LLMs
• InfiFPO: Implicit Model Fusion via Preference Optimization in Large Language Models
• Sequential Bayesian Evaluation of Large Language Model Behavior
• Budget-Aware Anytime Reasoning with LLM-Synthesized Preference Data

Source: arXiv cs.CL | 2026-10-15