The article explains that stuffing a RAG system with many overlapping or conflicting documents consumes tokens, slows responses, and introduces noise that prevents the model from correctly using the most relevant evidence, ultimately worsening answer quality.
The article explains why a plain similarity‑threshold check cannot reliably stop Retrieval‑Augmented Generation from fabricating answers, introduces a four‑stage evidence‑control framework, details how to calibrate thresholds with balanced positive and negative samples, and outlines concrete actions for handling insufficient evidence.
InstEmb introduces learnable look‑ahead tokens and dual‑anchor alignment pooling to inject future output semantics into embeddings, enabling instruction‑following tasks such as product retrieval to achieve strong performance without any extra decoding overhead.
The article breaks down Retrieval‑Augmented Generation into offline indexing (document chunking, embedding, vector storage) and online query (question embedding, similarity search, augmented generation), highlights retrieval accuracy as the main bottleneck, and shows practical use cases such as internal knowledge bases and product documentation assistants.
The article explains that choosing between vector databases and knowledge graphs for RAG systems depends on data structure and query style, compares their strengths and weaknesses with concrete examples, code snippets, and hybrid solutions, and offers practical guidance to avoid common pitfalls.
The article breaks down LlamaIndex RAG into three production stages—pre‑retrieval (metadata filtering, document constraints, query transformation), during retrieval (vector‑store ANN/hybrid/MMR, query modes, embedding consistency, extended retrievers) and post‑retrieval (filtering, reranking, context expansion, layout and recency handling)—and provides concrete code snippets, component choices, and practical recommendations for each stage.
This article explains how to use LangGraph to create a hybrid RAG agent that dynamically selects between vector, graph, web, or direct LLM retrieval, detailing the router, grader, rewriter, generator, and hallucination‑checking components along with a complete Python implementation.
Most retrieval systems apply the same pipeline to every query, but this one‑size‑fits‑all approach fails on hard queries and wastes compute on easy ones; the article defines cheap, no‑LLM signals to predict weak retrieval, evaluates them across three corpora, and proposes a gating method to upgrade only the weak cases.
The article analyzes the shift to agent‑centric AI, explains why traditional retrieval creates a costly "Ten blue links" loop, and details how Pinecone Nexus’s context compiler and composable retriever, together with the KnowQL language, provide structured, governed knowledge that boosts task completion rates, cuts latency, and reduces token usage by up to 90%.
The article examines why traditional single‑step Retrieval‑Augmented Generation fails when retrieved passages are irrelevant, outlines the three fundamental flaws of that pipeline, and presents the Agentic RAG paradigm—turning retrieval into a reusable tool with planning, reflection, and decision loops, illustrated with code, interview scenarios, and practical deployment tips.
The paper introduces Direct Corpus Interaction (DCI), letting LLM agents bypass vector indexes and use command‑line tools like grep to directly search raw text, achieving higher accuracy and lower cost on complex multi‑hop QA and retrieval benchmarks.
This article explains what Agentic RAG is, contrasts it with ordinary RAG by detailing its dynamic decision‑making, multi‑step retrieval loop, higher cost and latency, and suitable scenarios, and outlines two implementation patterns—single‑agent and multi‑agent—plus a concise interview response.
The article argues that memory for AI agents is not mere storage but a closed‑loop system comprising a raw ledger, derived views, and a policy layer, and examines how non‑parametric memory, time‑aware structures, and system‑2 control affect scalability, reliability, and performance.
Even with the world’s best large language models, chaotic internal notes, calls, and processes turn enterprise AI output into junk; a five‑layer architecture—capture, retrieval, source‑truth, permission, and feedback—plus a six‑question test can turn a noisy "company brain" into a useful tool, as shown by Single Grain’s dramatic time‑saving results.
In 2026, Retrieval‑Augmented Generation (RAG) has become a core reasoning paradigm for high‑compliance domains, prompting a shift from simple output correctness to multi‑stage falsifiable testing, dynamic adversarial knowledge graphs, LLM‑as‑Tester automation, and audit‑ready compliance reporting.
The article explains why a structured test set is essential for Retrieval‑Augmented Generation systems, outlines failure modes, describes layered evaluation of retrieval and generation, details infrastructure like chunk IDs and manifests, and provides a complete annotation pipeline with cold‑start and adversarial strategies.
When a long document is split into many highly similar chunks, vector‑based top‑k retrieval tends to return multiple pieces from the same source, causing document dominance; applying a per‑document chunk limit together with Maximal Marginal Relevance (MMR) re‑ranking introduces diversity while preserving relevance, offering a low‑cost way to improve RAG answer quality.
The INTRA paper shows that using a decoder’s cross‑attention as an internal retrieval mechanism eliminates the need for a separate retriever, achieving state‑of‑the‑art multihop QA performance with only 164 K trainable parameters and shared pre‑encoded representations.
This article breaks down the underlying logic, ideal scenarios, benchmark data, decision trees, and real‑world case studies for full‑text (BM25), vector, and graph retrieval, showing why hybrid approaches dominate production while each technique has distinct strengths and trade‑offs.
The article explains agentic search as an LLM‑driven, multi‑step retrieval process, contrasts it with traditional RAG pipelines, provides concrete examples, discusses when each approach is appropriate, and argues that agentic search will augment rather than fully replace RAG.
After achieving high recall with hybrid Elasticsearch and vector search, the article shows how inserting a reranker—either Cohere's cloud API or a local Cross‑Encoder—compresses the top‑20 candidates to the most relevant three to five, dramatically improving answer accuracy, cutting token costs, and detailing a dual‑track implementation for production and development environments.
The article explains how AI agents can acquire long‑term memory by combining three functions—coherence, context, and learning—with four memory types, describes the full retrieval‑store loop, and provides a step‑by‑step Python implementation using OpenAI embeddings, ChromaDB, and forgetting strategies.
The recent arXiv survey on Agent Skills maps the full lifecycle of skills—representation, acquisition, retrieval, and evolution—and warns that unchecked growth can turn a valuable process asset into technical debt, urging teams to enforce admission quality, robust routing, versioning, testing, and retirement mechanisms.
The article questions whether dedicated vector databases are truly needed for AI applications, examining market hype, the rapid emergence of many vector‑DB products, real‑world examples like PostgreSQL pgvector and major vendor integrations, and the hidden costs of data fragmentation and operational complexity.
The article examines why a model’s advertised context window (e.g., 128 K or 1 M tokens) does not guarantee effective long‑context reasoning, summarizing the RULER framework that breaks long‑context ability into retrieval, interference resistance, multi‑hop tracking, aggregation, and multi‑answer recall, and offering practical guidance for evaluating and using such models.
The article explains why stateless LLM calls require a structured memory system for AI agents, describes four memory types, a five‑stage pipeline, design patterns, common pitfalls, and provides a detailed production architecture with performance numbers and code examples.
The article dissects Claude Code’s memory system, showing that the real challenge is deciding what information to keep and when to discard, and it details design principles, index‑content separation, LLM‑based retrieval, expiration handling, write‑path isolation, and practical improvements applied to the author’s own agent platform.
The article analyses why a real agent must have memory, defining memory as an external state that feeds decision‑making, proposing a three‑part architecture (Raw Ledger, Views, Policy), contrasting parametric and non‑parametric approaches, and detailing bottlenecks, temporal handling, and procedural extensions.
The article explains why RAG systems require stage‑wise evaluation, detailing retrieval metrics such as Precision, Recall, F1, MRR, NDCG and Context Relevance, and generation metrics like Faithfulness, Answer Relevance and Completeness, while discussing LLM‑as‑Judge automation and a three‑layer assessment framework.
This guide outlines a comprehensive RAG evaluation framework covering failure modes, multi‑layer metrics, test‑set construction, open‑source tools, CI/CD quality gates, production monitoring, and special considerations for agentic RAG to ensure reliable, trustworthy retrieval‑augmented generation systems.
The article traces the evolution of Retrieval‑Augmented Generation from its 2020 Naive baseline through Advanced, Modular, Graph, and Agentic generations, detailing architectural shifts, optimization techniques, self‑correction mechanisms, and future challenges such as long‑context handling and multimodal retrieval.
Because LLMs have limited context windows, the article shows how to feed a 50‑page PDF to an AI by loading the document, splitting it into manageable chunks, converting each chunk into embeddings, storing them in a vector database, and then retrieving the most relevant passages for answering questions.
This article outlines five critical design principles for constructing high‑quality Retrieval‑Augmented Generation (RAG) systems, covering document chunking strategies, embedding model selection, hybrid retrieval architectures, metadata filtering with multi‑level indexes, and reranking mechanisms, and provides concrete code snippets and evaluation metrics.
This article dissects Claude Code's memory system, explains why unstructured memory degrades performance, introduces four distinct memory types with concrete examples and schema, shows how to handle expiration and retrieval strategies, and provides step‑by‑step implementation code to improve agent reliability.
After struggling with naive RAG that delivered only 60% accuracy, the author outlines eleven advanced strategies—including context-aware chunking, query expansion, re‑ranking, multi‑query, knowledge graphs, and agent‑based retrieval—that together raise performance to 94%, and provides detailed implementation examples, trade‑offs, and a step‑by‑step deployment roadmap.
The author explores Clawvard, an AI‑agent assessment platform that tests agents across eight dimensions, shares personal test results showing an initial A‑ rating with a critical retrieval weakness, details the customized improvement rules applied, and demonstrates a subsequent A+ rating, while also discussing the platform’s limits and practical use cases.
The article explains how to assess Retrieval-Augmented Generation (RAG) projects using the Ragas automated evaluation framework, detailing four key dimensions—recall quality, answer faithfulness, answer relevance, and context utilization—and describes the underlying metrics for both retrieval and generation stages.
The article explains how to engineer persistent, cross‑session memory for AI agents by persisting key user facts, task states, and decisions in a multi‑layer storage architecture, detailing retrieval before each request and update after each interaction.
This article analyzes Claude Code's memory system, explaining how typed storage separates user, feedback, project, and reference data, how an indexed MEMORY.md file keeps the index lightweight, how triggered retrieval balances relevance, freshness, and reliability, and why pre‑use validation prevents stale or incorrect facts from contaminating model responses.
This guide explains how to create a retrieval‑augmented generation (RAG) system that avoids embeddings by converting documents into a hierarchical tree, using an LLM to navigate, summarize, and retrieve answers, complete with a full Python implementation and a GitHub repository.
Retrieval‑Augmented Generation (RAG) lets large language models first fetch relevant documents and then generate answers, addressing the inability of models to answer private or domain‑specific queries by precisely feeding them the most pertinent knowledge.
This article explains the six pillars of context engineering, focusing on structuring techniques, advanced retrieval methods, hybrid search, reranking, query transformation, and practical pipelines that turn raw data into reliable, LLM‑ready inputs for higher quality AI responses.
Chunk splitting, a critical yet often overlooked step in RAG pipelines, dramatically impacts retrieval recall and LLM output quality; this guide walks through three evolution stages—from naive fixed‑length splits to sentence‑aware overlaps and finally semantic, structure‑driven segmentation—complete with code, experiments, and practical pitfalls.
This article outlines a full‑stack RAG architecture—offline parsing, query understanding, online retrieval, and context generation—explains six critical module interactions, and provides a concise interview framework for presenting the design from start to finish.
This tutorial walks beginners through Python fundamentals, LangChain installation, core components like prompt templates, chains, and retrieval, and culminates in a complete intelligent customer‑service chatbot, showing step‑by‑step code and practical tips for building AI applications.
MemSifter introduces a lightweight agent that outsources memory retrieval for large language models, using a Think‑and‑Rank pipeline and a task‑result‑oriented reinforcement‑learning training paradigm to achieve superior retrieval accuracy and efficiency across eight benchmark tasks while keeping inference overhead minimal.
This article explains why retrieval‑augmented generation (RAG) models often hallucinate, introduces the concept of context engineering, and details six practical techniques—including selective retrieval, context compression, hierarchical layout, dynamic query rewriting, memory management, and tool‑aware context—along with their trade‑offs and real‑world impact.
The article analyzes how six leading AI‑agent providers—Manus, Cursor, Anthropic, OpenAI, Google, and LangChain—tackle the fundamental problem of when and how a large language model should see information, detailing each solution, a cross‑company comparison matrix, consensus points, controversies, and open research questions.
This article details Tencent Search's exploration of large‑model‑driven ranking, covering the evolution from traditional keyword retrieval to RAG‑based AI search, the multi‑stage AI ranking architecture (L0‑L5), model training pipelines, distillation, synthetic data generation, and future research directions.
This article surveys recent advances in graph‑structured agent memory, presenting a taxonomy, lifecycle stages from extraction to evolution, open‑source tools, and benchmark suites that together illustrate how graph memory can overcome knowledge truncation, tool incompetence, and performance saturation in LLM‑driven AI agents.
This article presents the Trustworthy Generation design pattern for Retrieval‑Augmented Generation (RAG) systems, analyzes four root causes of low trustworthiness—retrieval errors, content reliability, pre‑retrieval reasoning mistakes, and model hallucinations—and proposes layered solutions, citation techniques, CRAG and Self‑RAG architectures, guardrails, and practical trade‑offs.
This article explains how to build a controllable, editable, and cost‑effective long‑term memory system for AI agents by categorizing memory types, structuring a three‑stage pipeline of capture, AI‑driven compression, and smart retrieval, and choosing appropriate storage back‑ends such as files, knowledge bases, or databases.
The article presents Agent KB, a universal memory infrastructure that lets heterogeneous AI agents share experiences through a Reason‑Retrieve‑Refine pipeline and a teacher‑student dual‑agent architecture, showing significant performance gains across benchmarks like GAIA, SWE‑bench, and various LLM families.
This article outlines five practical engineering strategies—multi‑vector retrieval, manual splitting and labeling, scalar enhancement, context augmentation, and dense‑sparse vector integration—that together address common RAG retrieval bottlenecks and dramatically improve recall stability and answer quality.
This article breaks down the memory systems behind LLM‑based agents, explaining why persistent memory is needed, the differences between short‑term context buffers and long‑term vector stores, practical implementation choices, maintenance strategies, and how to articulate these concepts effectively in technical interviews.
This article walks through the core RAG pipeline, explains why chunking is the linchpin of retrieval quality, and provides detailed definitions, trade‑offs, and implementation examples for five chunking techniques—fixed, recursive, semantic, structure‑aware, and delayed—so you can choose the right approach for any document‑heavy AI application.
This guide shows how to evolve a simple Retrieval‑Augmented Generation prototype into a production‑grade system by strengthening data ingestion, optimizing retrieval with hybrid and reranking techniques, adding query rewriting, long‑context handling, reinforcement learning, and multimodal support, so candidates can demonstrate real engineering depth in interviews.
This article explains how proper document chunking—choosing the right chunk size, overlap, and structure‑aware boundaries—directly impacts the relevance, factuality, and efficiency of Retrieval‑Augmented Generation pipelines, and provides multiple Python implementations ranging from simple fixed‑length splits to semantic and hybrid approaches.
This article explains the limitations of current pointwise in‑context learning methods, introduces a novel ranking‑based approach called SeDPO that learns preference orders among examples, and demonstrates its superior performance across multiple NLP tasks through extensive experiments and ablation studies.
This article explores the emerging concept of context engineering for AI agents, explains why managing long‑range context is critical, and details five practical strategies—Offload, Reduce, Retrieve, Isolate, and Cache—backed by insights from leading industry teams and the "Bitter Lesson" philosophy.
This article explains why initial retrieval in Retrieval‑Augmented Generation often yields noisy results, describes how rerankers act as quality filters to improve relevance, compares the leading 2025 reranker models—including Cohere, bge‑reranker, Voyage, Jina, FlashRank, and MixedBread—and provides code snippets, evaluation metrics, and guidance for selecting the right model for specific use cases.
This article explains the core architecture of AI agents powered by large language models, detailing how planning, short‑term and long‑term memory, and tool integration work together through vector databases, retrieval‑augmented generation, and summarization to enable stateful, intelligent interactions across multiple sessions.
This article introduces Pos2Distill, a novel knowledge‑distillation framework that transfers capabilities from advantageous to disadvantaged positions in large language models, effectively mitigating positional bias and improving performance on long‑text retrieval and in‑context reasoning tasks.
Jeff Huber, founder of Chroma, argues that traditional RAG is obsolete, introduces context engineering as the new paradigm, and shares five practical retrieval strategies, a complete pipeline, and insights on handling context rot, memory, and generative benchmarking to build production‑grade AI applications.
This article explores why proper document chunking is crucial for Retrieval‑Augmented Generation, explains core concepts like context windows and signal‑to‑noise, compares various chunking strategies—from simple fixed‑size splits to semantic and hybrid approaches—and provides practical Python code examples to help you build more effective RAG pipelines.
This article introduces ConfuseBench, a benchmark that classifies LLM uncertainty into document‑missing, ability‑limited, and ambiguous types, and presents methods—including retrieval, chain‑of‑thought, and clarification—to detect and actively resolve uncertainty, improving answer quality across diverse tasks.
This comprehensive guide walks you through the fundamentals of Retrieval‑Augmented Generation (RAG) and AI agents, explains their inner workings, shares optimization tricks, provides ready‑to‑run code snippets, and demonstrates how to evaluate performance with metrics such as recall, faithfulness, and answer relevance.
This article presents a step‑by‑step guide for constructing a scalable enterprise Retrieval‑Augmented Generation (RAG) solution using the Model Context Protocol (MCP), covering architecture comparison, system design, Milvus‑backed knowledge store, Python client implementation, deployment scripts, code examples, and best‑practice recommendations.
This article presents a comprehensive overview of Zhihu Direct Answer, describing its AI‑driven search architecture, RAG framework, query understanding, retrieval, chunking, reranking, generation, evaluation mechanisms, engineering optimizations, and the professional edition, while sharing concrete performance‑boosting practices and future development plans.
This study introduces DGDE, a document‑grounded dialogue framework that leverages large language model‑generated evidence, combining retrieval, reranking, fine‑tuning, and iterative question correction to markedly improve accuracy, comprehensiveness, coherence, and completeness on the Doc2dial benchmark.
This article analyzes common shortcomings of RAG pipelines—data preparation, retrieval, and LLM generation—and provides concrete optimization techniques such as advanced chunking, embedding model selection, retrieval parameter tuning, rerank models, and prompt engineering, promising up to a 20% performance gain.
Conan‑Embedding‑V2, a newly trained 1.4 B‑parameter LLM with a custom tokenizer, 32 k token context, SoftMask, cross‑lingual retrieval data and dynamic hard‑negative mining, delivers state‑of‑the‑art multilingual embeddings that surpass larger models on both English and Chinese MTEB benchmarks while remaining compact and fast.
This article breaks down Retrieval‑Augmented Generation (RAG) into its three core stages—data preparation, data retrieval, and LLM generation—showing how document chunking, embedding, vector databases, similarity search, and optional re‑ranking combine to let large language models produce more accurate, knowledge‑grounded answers.
Explore the BGE (BAAI General Embedding) family—including v1, v1.5, M3, Multilingual Gemma2, and EN‑ICL—detailing their multilingual capabilities, model variants, token limits, optimal use cases, and step‑by‑step installation and Python usage instructions with code examples for embedding generation and similarity scoring.
This article explains how to use the DB‑GPT application framework to improve Retrieval‑Augmented Generation (RAG) by detailing knowledge loading, chunking, embedding, graph extraction, storage options, retrieval strategies, workflow customization, evaluation metrics, and real‑world case studies.
This article dissects NVIDIA’s open‑source NV‑Embed model, explaining its decoder‑only architecture, latent attention layer, two‑stage contrastive training, data curation strategies, and experimental results that together push embedding performance to the top of the MTEB benchmark.
The article explains how the inherent knowledge‑staleness, hallucination, lack of private data, non‑traceable output, limited long‑text handling, and data‑security concerns of large language models can be mitigated by Retrieval‑Augmented Generation, which combines external retrieval, augmentation, and generation to provide up‑to‑date, reliable, and secure AI responses.
Retrieval‑Augmented Generation (RAG) using LangChain lets developers enhance large language models by embedding user queries, fetching relevant documents from a vector store, inserting the context into a prompt template, and generating concise, source‑grounded answers, offering low‑cost, up‑to‑date knowledge while reducing hallucinations and fine‑tuning expenses.
This article summarizes the key technical insights from Zhihu Direct Answer, an AI-powered search product, covering its product overview, RAG framework, query understanding, retrieval strategies, chunking, reranking, generation techniques, evaluation methods, and engineering optimizations for cost and performance.
This article describes a low‑cost, privacy‑preserving chatbot for logistics that combines data cleaning, large‑model‑based data augmentation, BM25 and vector retrieval, a DNN rerank model, and LLM‑driven answer rewriting to deliver accurate, compliant automated responses.
ChatDBA, developed by Shanghai Aikesheng, is an AI-driven database operation assistant that leverages large language models and Retrieval‑Augmented Generation to provide fault diagnosis, knowledge learning, SQL generation and optimization, addressing challenges such as vague outputs, complex troubleshooting logic, and memory management through a structured architecture and multi‑modal retrieval strategies.
This article reviews Wang et al.'s 2024 research on Retrieval‑Augmented Generation, outlining optimal practices such as query classification, chunk sizing, hybrid metadata search, embedding selection, vector databases, query transformation, reranking, document repacking, summarization, fine‑tuning, and multimodal retrieval to guide developers in constructing high‑performance RAG pipelines.
The KAG framework combines knowledge‑graph‑driven symbolic reasoning with large language model generation to improve accuracy, reduce hallucinations, and enable controllable, domain‑specific AI services such as government and medical Q&A, with open‑source support via OpenSPG and TuGraph‑DB.
This article reviews the latest vector database research showcased at SIGMOD 2024, covering system designs such as Starling, Vexless, RaBitQ, and ACORN, and discusses current academic hotspots including query processing, index structures, optimization techniques, and hardware acceleration for large‑scale similarity search.
An exclusive interview with Baidu’s senior database architects reveals the motivations behind building a dedicated enterprise vector database, details its novel column‑store engine, C++‑based retrieval stack, performance gains over open‑source solutions, multi‑modal support, RAG integration, and future research directions.
This article walks through the design and implementation of an Agentic Retrieval‑Augmented Generation system using LlamaIndex, showing how to wrap multiple RAG engines as tools, orchestrate them with hierarchical AI agents, and scale the solution with tool retrieval for large document collections.
This article examines the evolution of Retrieval‑Augmented Generation (RAG) architectures for large language models, outlines the challenges they face, introduces the modular RAG Flow concept with four workflow paradigms, and provides a step‑by‑step implementation using LangChain and LlamaIndex with code examples.
This article explains how Alibaba Cloud OpenSearch leverages vector retrieval, engineering and algorithmic optimizations, heterogeneous CPU‑GPU computing, and dense‑sparse hybrid memory to deliver billion‑scale, high‑throughput search performance and enable conversational AI use cases such as intelligent Q&A and SmartArXiv.
This article describes Kuaishou's approach to cold-start content recommendation, outlining the problems addressed, challenges in modeling sparse new videos, and solutions including graph neural networks, I2U retrieval, TDM hierarchical retrieval, bias correction, and future research directions.
This article explores how combining traditional keyword search with large language models (LLMs) enhances understanding of user intent, builds a robust automotive knowledge base, and delivers more accurate, context‑aware answers through a multi‑stage retrieval and generation pipeline.
This article explores how vector databases serve as the memory layer for large AI models, detailing their distributed, compute‑separated architecture, performance optimizations, hybrid vector‑scalar retrieval, and practical deployments across TikTok’s ecosystem such as image search, intelligent Q&A, and multimodal AI services.
A comprehensive look at how companies are rapidly adopting large language model APIs, retrieval techniques, and custom model strategies, revealing key statistics, emerging toolchains, and the shifting balance between closed‑source LLM services and open‑source custom stacks.
OpenAI has unveiled a suite of ChatGPT plugins—including a web‑browser, a code interpreter, a retrieval tool, and support for third‑party services—enabling the model to access up‑to‑date information, run Python code, query vector databases, and integrate external APIs, dramatically expanding its practical capabilities.
OpenAI announced a safety‑focused ChatGPT plugin system that connects the model to third‑party APIs for real‑time information retrieval, knowledge‑base access, and task execution, showcasing first‑party browser and code‑interpreter plugins, third‑party extensions, an open‑source retrieval plugin, and a detailed debate on security implications.
This article presents a comprehensive overview of industry search, covering its background, core retrieval and ranking technologies—including sparse and dense retrieval, pre‑trained language models, tokenization, NER, adaptive multi‑task training, and re‑ranking models—followed by detailed case studies such as address analysis, family‑ID unification, emergency call handling, education photo‑search, and power‑knowledge‑base integration.
This article presents a comprehensive retrieval‑based dialogue system designed to assist customer‑service agents by recommending candidate replies, detailing its five‑layer architecture, metric suite, text and vector retrieval modules, ranking strategies, and real‑world deployment results across multiple business scenarios.
The article details the architecture and techniques of ZhiZhuan's low-result search module, explaining how it combines ElasticSearch hard matching and sBert semantic vector soft matching, along with sophisticated negative sample strategies, to improve recommendation coverage and user experience.
This article reviews recent research on e‑commerce search, covering transformer‑based complementary rankings, Alibaba's cognitive concept net and its extension, joint deep retrieval with product quantization, personalized semantic retrieval, multi‑granularity deep semantic retrieval, and graph‑attention networks for long‑tail shop search.
The article introduces the Dual Vector Foil (DVF) algorithm system, which decouples index construction from model training to enable lightweight, high‑precision large‑scale recall using arbitrary complex models, and details its two‑stage and one‑stage solutions, graph‑based retrieval implementation, performance optimizations, and experimental results.
This article explains Alibaba's Deep Tree Matching (TDM) technology, covering the challenges of large‑scale match retrieval, the progression from classic two‑stage recall to tree‑based indexing, max‑heap tree modeling, beam‑search retrieval, and the joint model‑index learning across TDM 1.0, 2.0, and 3.0, highlighting significant offline and online performance gains and future research directions.
