Next point‑of‑interest (POI) recommendation aims to predict a user's next destination from spatio‑temporal check‑in data and underpins many location‑based services. Recent work incorporates large language models (LLMs) by textualizing user trajectories to exploit their semantic reasoning, yet directly generating POI identifiers with LLMs is unreliable because POIs are sparse, localized, and long‑tailed entities with weak natural‑language semantics, creating a semantic‑to‑entity gap between high‑level mobility intent and concrete POI prediction. To bridge this gap, we propose the ANCHOR framework for semantic‑to‑entity grounding. ANCHOR first estimates category‑level mobility intent in a compact, stable semantic space instead of predicting concrete POIs. The inferred intent is then grounded into a candidate set by integrating three complementary sources of mobility evidence: category anchors, historical visit memory, and user preferences. Finally, a history‑aware reranker selects the target POI by matching each candidate with the user's historical mobility patterns and current context. Experiments on three real‑world check‑in datasets show that ANCHOR consistently outperforms representative traditional, structure‑enhanced, and LLM‑based baselines across reported metrics. Further analyses validate the effectiveness of semantic‑to‑entity grounding and demonstrate favorable trade‑offs among candidate coverage, ranking quality, and inference efficiency.
Review: ANCHOR's two‑stage design—first estimating category intent then grounding to entities—effectively mitigates POI semantic sparsity, balancing recommendation accuracy with computational efficiency and offering a practical pathway for deploying LLMs in location recommendation.