Alternative property recommendations play a critical role in vacation rental marketplaces, helping users discover relevant options when viewing a specific listing. However, generating high-quality candidate alternatives presents unique challenges: heterogeneous inventory, geographic constraints, rapid availability changes, and long-tail property distributions. We present a comprehensive study of candidate generation (CG) approaches for vacation rental alternatives, comparing collaborative filtering, shallow embeddings, and graph neural network (GNN) methods. Our experiments on a large-scale vacation rental platform (over 2M active properties) show that a hybrid architecture combining item-based collaborative filtering with GNN-based retrieval improves Recall@300 by 14.8% over the strongest baseline, by leveraging the complementary strengths of the two sources: collaborative filtering excels at early recall for properties with rich interaction history, while GNNs discover diverse, non-obvious alternatives and handle cold-start scenarios more effectively. As a component result, GNN-based embeddings alone substantially outperform shallow Hotel2Vec embeddings (48-68% relative recall improvement across K), motivating their inclusion in the ensemble. Crucially, we examine how CG-stage gains carry through to the downstream ranking stage, and find that a stronger candidate pool yields higher downstream ranking quality, though attributing this effect cleanly is complicated by the coupling between candidate generation and ranker training. This recall-conversion gap is an important consideration for practitioners deploying new retrieval methods in two-stage recommendation systems.
Figures & tables
Figure 1: Alternative property recommendations on a vacation rental platform. When a user views a reference property Pref , the system generates a carousel of K alternative properties {P1,P2,...,PK} that match the user’s destination and preferences.
Figure 2: Heterogeneous graph schema for GNN training. Properties are split into LHS (reference) and RHS (candidate) nodes to learn asymmetric relationships.
Table 1: Comparison of candidate generation approaches.
Figure 3: Multi-source candidate generation architecture. CG sources (IBKNN, GNN) generate candidates independently. Business rules (including geo-filtering) are applied, candidates are merged, and lightweight re-ranking produces the final candidate set for the Final Re-Ranker.
Figure 4: Recall@K comparison between Hotel2Vec baseline and GNN. GNN achieves 48–68% relative improvement across all K values, with largest gains at mid-range (K=50, 100).
Table 3: Candidate-generation NDCG@ K for the same configurations and runs as Table 2 . Recall (Table 2 ) is the primary candidate-generation metric; NDCG@ K is reported alongside it to characterize within-pool placement.
Source
R@10
R@50
R@100
R@200
R@300
R@500
IBKNN
7.6%
20.6%
28.5%
37.2%
42.4%
48.5%
GNN
4.9%
16.6%
25.0%
35.2%
41.4%
49.2%
Hybrid (fused)
7.6%
21.2%
29.5%
38.8%
44.3%
51.0%
Table 4: Per-source candidate-generation Recall@ K (retrieval stage, before lightweight re-ranking) on the June 2025 test set. IBKNN leads at early K ; the GNN overtakes at high K ; the fused hybrid matches or exceeds both, with a clear margin from K=50 onward.
Candidate Pool
Test N@5
Val N@5
Baseline (IBKNN + Hotel2Vec)
0.345
0.361
Hybrid, proposed (IBKNN + GNN)
0.365
0.369
Table 5: Downstream ranking quality by candidate pool. Both rows use the same Final Re-Ranker with an identical feature set and architecture, trained and evaluated on its own candidate pool.
Vacation rental marketplaces face a structural imbalance on the supply side: a small fraction of properties receive most user interactions, while the long tail of new, niche, and seasonal listings generates too little behavioral signal for collaborative filtering to serve effectively. At Vrbo, item-based k-nearest neighbors (IBKNN) is a core candidate generation channel, but leaves tens of thousands of properties with no candidates and produces weak neighborhoods for sparsely interacted ones. We present a training-free, LLM-based candidate generation pipeline that complements IBKNN using static property metadata alone. An off-the-shelf LLM synthesizes diverse semantic queries per property, a pre-trained text encoder embeds them, and an approximate nearest-neighbor index retrieves candidates from an 11.7M-property catalog. A Union fusion strategy merges these with IBKNN while preserving the behavioral channel's ordering, guaranteeing no degradation on well-served properties, and a downstream learning-to-rank model re-scores the fused pool. Evaluated on 1.6M focal properties, the system extends candidate coverage to tens of thousands of properties IBKNN cannot reach, delivers its largest gains on the long-tail segment where behavioral methods are weakest, and matches or beats IBKNN at every K on shared properties. A downstream learning-to-rank stage further lifts the fused pool, yielding a complete candidate generation and re-ranking stack that serves the long tail without regressing well-served properties. We additionally show that Union fusion collapses the recall gap between a 3B open-weights LLM and frontier API-based models from 27-46% to under 1%, supporting self-hosted small-model deployment at marketplace catalog scale.
Syed Mohammed Arshad Zaidi, Eric Rincon, Shayan Hassantabar
Candidate-dependent routing lets a sequential recommender weight a user's past behaviors differently for each item it scores. We test whether this flexibility helps by adding a small residual routing module to a self-attentive baseline, in two versions that differ only in what sets the routing weights: the user's history representation or the candidate item's embedding. Across five paired seeds on the REES46 cosmetics-shop dataset, with matched training for all three models, candidate-dependent routing leaves the primary metric, normalized discounted cumulative gain at cutoff ten (NDCG@10), essentially unchanged (mean relative change +0.003%). Secondary and exploratory analyses show that this flat average hides two effects that cancel. The candidate-dependent model places more relevant items in its top ten (recall +1.81%) but ranks the first one lower (mean reciprocal rank -1.09%), and most of its new hits land at positions six to ten. Its gains come from queries with at least one target among the user's last 50 events, while queries whose targets all lie outside that window get worse. In this form, candidate-dependent routing trades accuracy at the top of the list for recall further down.
Han Chen, Yingrui Li
Department of Computer Science, The University of Texas at Austin, Austin, Texas, United States · Unaffiliated, Fairfax, Virginia, United States
First-stage recommenders in multi-stage systems produce a ranked candidate list from which a limited prefix is forwarded to downstream rankers. Because each forwarded item must be processed by more expensive ranking stages, this shortlist cannot be arbitrarily large. The first-stage objective is therefore high coverage of relevant items within the forwarded prefix, commonly measured by Recall@k. A relevant item may be available deeper in the retrieved list but absent from the shorter prefix that is actually consumed. This paper studies post-hoc verification for promoting such candidates into the consumed shortlist without retraining or replacing the retriever. We introduce a lightweight generative verifier for retrieval models. Given a retriever state and a candidate item, the verifier scores the item through the likelihood of its identifier tokens. It is trained post hoc with next-token cross entropy, requires no sampled negatives or candidate pool during training, and scores only the retriever's top-K candidates at inference. The interface is minimal: the retriever supplies a query state and candidate items, and the item representation can use any fixed tokenization. Across Amazon product recommendation and YaMBDa music recommendation, the same verifier training recipe improves Recall@10 for SASRec, GRU4Rec, NextItNet, and MiniOneRec. Ablations show that the improvements are not explained solely by injecting item-content features into the retriever, supporting verification as a post-hoc output-side adaptation mechanism.