Aug 2026· Nexus Science Review· 0 citations· 19 references
TL;DR
LSR retains a general encoder as a transferable semantic substrate while explicit modules reorganize, augment, or index its representations for a defined operational objective.
Abstract
Dense embeddings are foundational to contemporary natural language processing, information retrieval, recommendation, and retrieval-augmented generation. Nevertheless, a single general-purpose vector typically superimposes multiple relations—topic, entailment, sentiment, part–whole structure, evidential role, temporality, and domain-specific constraints—within one geometry and one similarity function. This article presents layered semantic refinement (LSR) as a framework and evaluation protocol rather than as a single universally validated algorithm. LSR retains a general encoder as a transferable semantic substrate while explicit modules reorganize, augment, or index its representations for a defined operational objective. The framework covers learned projections, nonlinear adapters, graph message passing, hierarchical aggregation, deterministic rule channels, clustering, multi-view representations, and prompt-conditioned embeddings.
Deep sequential encoders such as Transformers are widely used for modeling user behavior in sequential recommendation, yet their effectiveness can be limited when interaction sequences are short and sparse. In such settings, insufficient behavioral evidence makes it difficult to learn reliable representations. Existing contrastive learning methods provide useful auxiliary supervision, but they often rely on random sequence perturbations that may remove informative interactions and typically apply contrastive objectives only to the final encoder layer. Moreover, interaction-based models cannot fully capture global semantic relations among items that rarely or never co-occur. To address these limitations, we propose Semantic Global Graph with Multi-layer Alignment for Recommendation (SeGMA4Rec), which combines item-level semantic enrichment with sequence-level hierarchical regularization. SeGMA4Rec constructs a semantic item graph from pretrained textual embeddings, adaptively refines the retained edge weights, and propagates task-relevant information among related items. The graph-enhanced item representations are then processed by a sequential encoder whose attention scores guide probabilistic view generation toward retaining informative interactions. Three contrastive objectives further align representations across augmented views and encoder layers, providing direct regularization at multiple depths. Experiments on four benchmark datasets show that SeGMA4Rec achieves strong and generally superior performance over competitive baselines, including improvements of up to 9.9% in Recall@10 and 8.1% in NDCG@10 on ML-1M. Additional analyses confirm the complementary contributions of semantic graph propagation, adaptive edge refinement, attention-guided augmentation, and multi-layer contrastive alignment.
Recent advances in neural topic models with pre-trained language models (PLMs) have achieved strong performance by leveraging general-domain pre-training, yet their topic interpretability often degrades on specialized corpora. This limitation primarily stems from the geometry of the embedding space, where domain-specific terms unseen during pre-training collapse into an indistinguishable region, and neither domain-specific re-training, word-level graph enrichment, nor parameter-efficient fine-tuning can restructure this space without inheriting the capacity ceiling of the underlying encoder. Our key insight is that a learnable graph layer operating on token-level PLM embeddings can acquire corpus-specific semantic structure that the frozen encoder lacks, because token-level graphs preserve document-local context that word-level representations discard and joint optimization with the topic objective reshapes embedding geometry directly from target-domain evidence. We instantiate this insight as DARTopic, a domain-agnostic framework that constructs token-level semantic graphs from frozen PLM embeddings and jointly trains a GNN encoder with topic inference. Across three benchmarks spanning general, biomedical, and legal domains, DARTopic consistently outperforms strong baselines in topic coherence and document clus- tering without any encoder fine-tuning, while demonstrating robustness to PLM choice and favorable runtime efficiency over fine-tuning based alternatives.
Composed video retrieval (CoVR) searches a gallery for the target video that realizes a natural-language modification of a source clip. However, at gallery scale, this creates a fundamental tension: compact embeddings enable efficient, reusable search but can miss the transient actions, state changes, and subtle constraints that demand fine-grained video reasoning, whereas applying large multimodal models uniformly sacrifices scalability. To address these limitations, we propose that frozen foundation models should instead occupy complementary roles, with inference depth adapted to query difficulty. Based on this premise, we introduce \methodname{}, a framework for training-free \methodexpansion{}. Specifically, a composed-query embedding first searches reusable video-only gallery representations; uncertain queries undergo bounded reranking and candidate expansion; ambiguous edits trigger target-description generation; and only close leading candidates reach multimodal verification. To support these roles, frame selection, spatial resolution, and time cues are adapted to each stage. Across complete target-gallery evaluations, our method reaches state-of-the-art performance among training-free approaches, with 89.55 and 93.43 R@1 on Dense-WebVid-CoVR and CoVR-R, respectively (with more than +35\% and +25\% absolute margins to the closest counterpart). These results show that adaptively orchestrating foundation-model capabilities can combine scalable retrieval with fine-grained reasoning without task-specific training. The source code and all relevant guidelines are available on https://github.com/demidovd98/CoVRAGE.
Dmitry Demidov, Muhammad Zaigham Zaheer, Omkar Thawakar et al.· 0 citations
Data sparsity remains a major obstacle for recommender systems because conventional collaborative filtering methods rely heavily on observed user–item interactions and often fail to exploit richer semantic and relational signals. To address this limitation, we propose SeRel-LightFM, a knowledge-aware hybrid recommendation framework built on top of LightFM. Our method combines two complementary feature-construction branches. First, we encode multi-field movie text with a Sentence-Transformer encoder and convert the resulting dense text embeddings into sparse semantic prototype features through K-Means clustering and soft assignment. Second, we construct a unified knowledge graph over users, movies, and metadata entities, learn joint embeddings with TransE, and transform these dense relational representations into sparse relational prototype features for both items and users. These newly constructed features, referred to as Advanced Relational and Semantic Representations (ARSR), are integrated with the original dataset-derived Conventional Feature Space (CFS) inside LightFM. Experiments on the multi-source CSP movie dataset under a leave-one-out evaluation protocol show that SeRel-LightFM consistently outperforms popularity, ItemKNN, and the LightFM baseline across Recall@K and NDCG@K at K = 5, 10, 20, and 50. The results indicate that combining semantic and relational prototype features improves ranking quality in sparse recommendation settings. The source code and dataset are available at https://github.com/gminh715/SeRel-LightFM.
M. Pham, V. Vu, Hung-Nghiep Tran· International Conference on...· 0 citations
Large Language Models (LLMs) have demonstrated remarkable general-purpose abilities across a wide range of domains, and these strengths have also been increasingly evidenced in recommender systems. However, existing methods that attempt to integrate collaborative signals into LLMs often fail to preserve their foundational knowledge. This loss is critical in text-rich recommendation, where robust semantic understanding is required to interpret user reviews and item profiles. We propose PALRec, a parameter-preserving augmentation framework that equips an LLM with recommendation capabilities while keeping its original parameters fixed. We first construct evidence-grounded user and item profiles from reviews and use them as concise pseudo-labels for reconstruction. We then introduce lightweight, trainable user and item embedding modules optimized with a multi-task objective that combines next-item prediction and profile reconstruction. These modules are trained jointly to align collaborative signals with the LLM’s semantic space without modifying the backbone. We also employ token-aware loss decomposition and frequency-aware reweighting to stabilize training and mitigate popularity bias. Experiments on public benchmarks show that PALRec consistently outperforms fully fine-tuned counterparts in recommendation accuracy while preserving the LLM’s pre-trained knowledge. This result highlights that maintaining the LLM’s semantic understanding is crucial for effectively exploiting textual information in recommender systems.
Hyunsoo Na, Minseok Gang, Sang-goo Lee et al.· ACM Transactions on Informat...· 0 citations
Volume-based multimodal retrieval jointly scores a text query with a candidate's video, audio, and subtitle embeddings. While this approach captures higher-order within-candidate alignment, the score remains candidate-local, and semantically related training samples primarily serve as contrastive negatives. This work introduces Hypergraph-Regularized Gramian Volumes (HyVol), a training-time module that incorporates these semantic relations prior to evaluating the original volume loss. Document hyperedges connect the observed modalities of each candidate, whereas semantic hyperedges link candidates whose detached captions are mutual top-k neighbors. A shallow gated hyper-graph network applies residual corrections to the modality embeddings. Presence masks exclude unavailable streams from message passing, and identity padding preserves the determinant of the observed Gram submatrix without feature imputation. As refinement operates on embeddings rather than scores, the same construction applies to both Gram and HyperGram. We remove the hypergraph after training, leaving the backbone-only architecture, original scoring function, and retrieval cost unchanged. We train both backbones on a 150K-clip subset of VAST-27M and evaluate zero-shot performance on six benchmarks. Under the paired protocol, HyVol improves R@1 across all five retrieval benchmarks, with video-to-text gains reaching +8.3 on MSR-VTT and +7.6 on VATEX. Under missing-modality masking, the V2T margin remains positive in all experimental settings, although the T2V margin becomes slightly negative in four.
