weaviate

Traditional vector retrieval relies on approximate nearest neighbor (ANN) search, which often yields semantically redundant results and fails to meet the diversity and contextual richness requirements of applications such as retrieval-augmented generation (RAG) and multi-hop question answering. To address this, we propose a novel paradigm—“Semantic Compression and Graph-Enhanced Retrieval”—that introduces submodular optimization into vector retrieval for the first time. We formalize semantic compression to maximize information coverage while explicitly suppressing redundancy. Leveraging information-geometric similarity metrics and k-nearest neighbor (kNN) graphs, we construct a multi-hop semantic search framework, further augmented with knowledge graph integration for structured semantic querying. Our method supports hybrid indexing and significantly improves semantic diversity and coverage in high-dimensional embedding spaces, outperforming state-of-the-art ANN baselines. The implementation is open-sourced, advancing research toward semantics-centric vector search.

This work proposes a bidirectional synergy framework between artificial intelligence and vector search to enhance semantic retrieval accuracy and mitigate knowledge obsolescence and hallucination in large language models (LLMs). The framework integrates “AI for Vector Search” (AI4VS), which optimizes learned indexes, adaptive pruning, and automated hyperparameter tuning, with “Vector Search for AI” (VS4AI), which strengthens retrieval-augmented generation (RAG) to improve LLM factuality and timeliness. By enabling end-to-end joint optimization, the approach significantly boosts both the semantic precision and efficiency of vector search while enhancing the knowledge currency and reliability of LLM-generated outputs. This paradigm offers a novel foundation for intelligent information systems that demand accurate, up-to-date, and trustworthy responses.

This work addresses the challenge that existing retrieval-augmented generation systems struggle to effectively aggregate dispersed evidence from multiple sources when handling complex queries, while approaches relying on explicit knowledge graphs suffer from high construction costs and poor compatibility. To overcome these limitations, the authors propose a graph-structured augmentation and reranking method that avoids building a full knowledge graph. During offline preprocessing, data objects are enriched with graph-based contextual information; at inference time, candidate results are reranked using graph-aware proximity measures. The approach is retrieval-agnostic, seamlessly integrates with mainstream vector databases, and significantly improves retrieval performance across multiple benchmarks—while maintaining low inference latency and strong system compatibility.

This work addresses the significant degradation in retrieval efficiency caused by fragmented connectivity in traditional graph indexes when handling low-selectivity filtered queries, where qualifying vectors are sparse. To overcome this limitation, the authors propose Curator, a partitioned dual-index architecture based on shared clustering trees that constructs dedicated sub-indexes for different labels, thereby maintaining high search efficiency while reducing memory overhead. Curator introduces an adaptive partitioning mechanism that supports incremental updates and enables on-the-fly temporary index construction during query execution, effectively accommodating complex predicate filtering. Experimental results demonstrate that, when integrated with state-of-the-art graph indexes, Curator reduces query latency by up to 20.9× for low-selectivity queries, with only a 5.5% increase in index construction time and a 4.3% increase in memory usage.

To address the challenges of semantic understanding and interactive analysis in open academic big data exploration, this paper pioneers the modeling of knowledge graph (KG) embedding spaces as interpretable semantic vector spaces—extending beyond their conventional use solely for link prediction. We propose a novel vector-algebraic paradigm for semantic querying—including analogical reasoning and similarity-based retrieval—that systematically uncovers structured semantic regularities among entities and relations in the embedding space. Our method integrates state-of-the-art KG embedding models (e.g., TransE, ComplEx) with word-embedding-style analogy analysis techniques to construct an explainable query framework tailored for scholarly KGs. Extensive evaluation on multiple public academic KGs demonstrates substantial improvements in deep relational pattern mining for entities such as papers and authors, and effectively supports cross-domain analogical inference and interactive exploratory tasks.

Traditional vector retrieval struggles to handle complex queries requiring structured reasoning in industrial knowledge graphs. This work constructs an aerospace supply chain knowledge graph comprising 46 node types and 64 relation types, and introduces the “operator vocabulary” hypothesis, positing that the bottleneck in graph reasoning lies not in model intelligence but in the availability of computational primitives. Guided by this insight, the authors design an LLM-driven query planner that integrates nine graph traversal primitives and six graph computation tools, forming a structure-aware retrieval-augmented generation framework. Evaluated on 23 queries spanning ten intent categories, the approach achieves an F1 score of 0.632, significantly outperforming a customized processor (0.472), and further exposes a systematic bias in existing entity-level F1 metrics when assessing structured queries.

Traditional vector retrieval relies on pairwise geometric similarity, which struggles to simultaneously achieve semantic alignment and consistency with the head-tail distribution of data. This work proposes a Graph Wiring framework combined with Spectral Indexing, modeling the embedding space as an energy network induced by the topology of feature column vectors. By integrating geometric similarity with spectral structural information and introducing τ-modulation for adaptive retrieval, the method leverages spectral graph theory, energy-based modeling, and epiplexity analysis. Implemented using the open-source arrowspace library, it significantly outperforms purely geometric retrieval across multiple benchmarks and industrial applications, effectively enhancing both semantic alignment and distributional consistency to meet the demands of modern RAG systems for flexible and efficient retrieval.

This study addresses the challenge of hyperparameter tuning for Hierarchical Navigable Small World (HNSW) graphs, which is complicated by nonlinear parameter couplings and stringent resource and time constraints in production environments. The work reveals, for the first time, that the HNSW configuration space exhibits structural properties—namely monotonicity, dominant unimodality, and separability—and leverages these insights to propose CHAT, a constraint-aware deterministic tuning framework. By integrating structure-guided pre-pruning and directed search, CHAT overcomes the efficiency limitations of conventional black-box optimization. Empirical evaluations across multiple datasets and vector engines demonstrate that CHAT achieves up to a 45% improvement in throughput or an 11% gain in recall compared to strong baselines, while accelerating tuning convergence by as much as 44×.

Existing search-augmented large reasoning models struggle to effectively leverage the topological information of graph structures, leading to inefficient retrieval and reasoning in zero-shot graph learning tasks. This work proposes GraphSearch, a novel framework that extends search-augmented reasoning to graph learning for the first time. GraphSearch employs a graph-aware query planner and a graph-aware retriever that jointly decouple semantic queries from the graph search space, introducing a hybrid retrieval mechanism that integrates both topological and semantic information. The framework supports both recursive and flexible graph traversal strategies, enabling efficient zero-shot inference without task-specific fine-tuning. Evaluated across multiple benchmark datasets, GraphSearch achieves state-of-the-art performance in zero-shot node classification and link prediction, matching or surpassing supervised methods.

This work addresses the limitations of existing methods in multi-vector similarity search, which rely on single-vector indexing and neglect inter-vector relationships, thereby struggling to balance recall and efficiency. The paper proposes MV-HNSW, the first native hierarchical navigable small-world graph index tailored for multi-vector data, enabling efficient and accurate retrieval by explicitly modeling vector dependencies. Key innovations include a novel edge weighting function that ensures symmetry, cardinality robustness, and query consistency, as well as a dynamic search strategy that enhances discovery of topologically isolated yet semantically relevant candidates. Extensive experiments on seven real-world datasets demonstrate that MV-HNSW achieves state-of-the-art performance, reducing search latency by up to 14× while maintaining recall above 90%.