This article explains what vectors are, how embedding models turn text into high‑dimensional vectors, the common distance metrics, why specialized vector databases are needed, and walks through a hands‑on Python demo with ChromaDB before covering production‑grade options, advanced tuning, and typical pitfalls.
JD’s Oxygen AIIC combines human‑in‑the‑loop ontology engineering, a semantic search‑then‑discrimination pipeline, and a self‑evolving multi‑task LLM/VLM model to produce high‑quality product knowledge for over a hundred thousand categories and billions of daily SKU updates, boosting search coverage to 80%, attribute auto‑fill to over 80%, cutting quality issues by 37% and raising click‑through by 9% while achieving 94.2% precision and 82.8% recall.
Open Notebook is an open‑source, fully private AI note‑taking assistant that consolidates PDFs, videos, web pages, and audio into searchable knowledge assets and can instantly generate multi‑speaker AI podcasts, offering a secure alternative to Google NotebookLM.
The article dissects Cursor's code‑base indexing pipeline, explaining how semantic vector search, trigram‑based regex filtering, AST‑driven chunking, custom embeddings trained on agent trajectories, Merkle‑tree change detection, and Turbopuffer's namespace‑per‑repo vector store combine to deliver sub‑second, accurate code retrieval even in monorepos with tens of thousands of files.
The article explains why RAG systems cannot rely on MySQL for embedding storage, shows the O(n) brute‑force search latency for hundreds of thousands of chunks, and demonstrates how vector databases with ANN indexes such as HNSW or IVFFLAT provide millisecond‑level response, high recall, and scalable storage.
This article explains why traditional keyword search fails on synonyms, introduces Spring AI's local ONNX‑based EmbeddingModel to convert text into high‑dimensional vectors, shows how to store and query those vectors with SimpleVectorStore, and compares production‑grade vector store options.
By integrating Milvus vector database with a Retrieval‑Augmented Generation architecture, the authors built an AI‑driven multi‑turn medical Q&A system that achieved 92% knowledge retrieval accuracy, 96.3% recall with 18 ms latency, and demonstrated the strengths, trade‑offs, and engineering practices of vector‑based semantic search in healthcare.
A vector database stores high‑dimensional numeric embeddings instead of traditional rows, enabling semantic similarity search through specialized indexes, metadata filtering, and massive scalability, while also presenting trade‑offs such as approximate results, update costs, and high memory consumption.
The article examines six practical challenges of AI‑First development—oversized tool libraries, when to trigger refactoring, propagating newly extracted methods, duplicate code from parallel sub‑agents, context aging, and the lack of a unified framework—while presenting concrete solutions such as three‑layer loading, sub‑agent isolation, semantic search, consolidation agents, persistent context files, and adaptive compression strategies.
The article reviews the three‑step RAG pipeline, explains why retrieval quality hinges on fast, accurate semantic matching, contrasts Bi-encoder’s offline vector indexing and speed with Cross-encoder’s token‑level interaction and higher precision, and discusses hybrid solutions such as ColBERT and LLM rerankers with practical engineering guidelines.
Adding dozens of MCP tools overloads the LLM’s context window, causing the “lost in the middle” effect that degrades accuracy, but a gateway with semantic tool discovery, role‑based virtual servers, and pre‑filtering can restore performance while preserving governance.
JoyCastle faced a growing ad‑asset library that slowed creative production, so it built an AI‑powered management system using Amazon Nova Multimodal Embeddings, achieving unified semantic search, automatic video segmentation, 96.7% recall and a 73.3% top‑2 precision while reducing manual labeling effort.
The article examines why naïvely feeding all dialogue history to large language models is costly and unreliable, and it walks through rolling context windows, inverted‑index pruning, semantic vector search, and GraphRAG as complementary techniques for creating efficient, reasoning‑capable AI agent memory.
The article explains why traditional keyword search often fails when user questions differ from knowledge‑base wording, introduces semantic search that matches queries and documents via vector similarity, details query understanding and rewriting techniques, lists common pitfalls, provides a full Python implementation, and shares best‑practice recommendations.
MemPalace is an open‑source, local‑first AI memory library that stores raw conversation and project content without summarisation, uses a hierarchical "memory palace" structure for fast semantic retrieval, provides plug‑in retrieval back‑ends, knowledge‑graph support, and achieves the highest publicly reported offline benchmark scores.
This article walks through deploying Milvus with Docker Compose, creating a persistent collection, tuning HNSW indexes, integrating LangChain.js for semantic retrieval, and covering performance tips and common pitfalls to run a production‑grade vector database.
The article presents a generative retrieval framework that anchors on a product‑side explicit semantic space, uses single‑sided query alignment, multi‑stage LLM‑driven attribute extraction, schema standardization, and 14B model SFT + DPO training to achieve interpretable, controllable, and high‑performance e‑commerce search.
GBrain, an open‑source AI memory hub created by YC partner Garry Tan, combines Postgres tsvector keyword search with pgvector semantic search via RRF, manages thousands of Markdown notes, and runs an automated nightly agent that refines and links memories, offering a practical long‑term recall layer for agents like OpenClaw and Hermes.
The article analyzes three leading AI experts' approaches to personal knowledge management—Karpathy’s five‑module LLM pipeline, Lex Fridman’s interactive voice‑driven consumption, and kepano’s cautionary separation of AI‑generated content—while detailing the author’s own downstream content‑production workflow that turns raw material into articles, videos, and social posts.
This article breaks down agentic memory into four distinct types—In‑context, External, Episodic, and Semantic/Parametric—explains three forgetting strategies (time decay, importance scoring, periodic consolidation), shows how memory flows through an agent loop, and provides complete Python code using OpenAI embeddings and ChromaDB for a production‑ready memory layer.
WeKnora is a Tencent‑open‑source LLM‑based document understanding and semantic search framework that extracts structured content from PDFs, Word files and images, offers agent‑driven reasoning, multi‑modal retrieval, and a modular architecture, with step‑by‑step Docker deployment and a web UI for instant querying.
This article walks through building a practical Elasticsearch demo that lets users type a single query like “nearby clean coffee shop” and get results by combining dense‑vector semantic search, geo filtering, BM25, and a hybrid RRF‑style ranking, with both LLM‑based structuring and a fallback hash‑based embedding.
Jina-grep is an open-source CLI that adds fast, MLX-powered semantic search to grep on Apple Silicon, offering three modes, sub-millisecond latency, high token throughput, and easy installation, making local code and log searching more accurate than keyword matching.
Moltbook, the AI‑only community formerly known as Motlbot, now hosts over 140,000 agents, 12,000 sub‑communities and tens of thousands of posts, while enforcing API‑key authentication, rate‑limit controls, heartbeat scheduling and semantic search, sparking debates about emergent AI behavior and safety.
Moltbook is a pioneering AI‑only social platform that lets agents post, comment and like, built with a Next.js front‑end, RESTful API back‑end, semantic vector search, heartbeat checks, dual human/agent modes and strict rate‑limiting to encourage quality over quantity while addressing scalability and security challenges.
Clawdbot, an open‑source AI assistant, uses local Markdown files and a SQLite‑based vector index to provide a transparent, searchable, and long‑term memory that separates temporary context from durable storage, enabling autonomous task handling across sessions.
WeKnora is an open‑source LLM‑driven framework that transforms complex, multi‑format documents into searchable semantic knowledge, offering features such as Agent mode, hybrid retrieval, secure private deployment, and an easy‑to‑use web UI, with step‑by‑step installation instructions and demo screenshots.
This article explains the challenges of managing hundreds of tools for AI agents, introduces a semantic‑driven tool selection approach via the Higress AI Gateway extension for AgentScope Java, and provides performance data, code examples, and a quick‑start guide for developers.
This article explains the principle of vector embeddings, shows how they turn words, images, audio and other data into dense numeric vectors, compares them with one‑hot encoding, describes static and contextual models, training methods, similarity metrics, and a wide range of real‑world AI applications.
This article explains the motivations, techniques, and practical implementations of code indexing—covering semantic vector‑based RAG pipelines and AST‑based structural analysis—to improve code navigation, AI‑assisted queries, security scanning, and development efficiency.
This article explains how next‑generation AI agents, combined with OpenSearch and the Model Context Protocol (MCP), enable semantic, multimodal, conversational, and agentic search, detailing the architecture, deployment options, toolset, and practical demos for sales analytics and DevOps use cases.
The article examines why file system access is crucial for autonomous agents, outlining common context‑engineering failures such as missing, excessive, or irrelevant information, and demonstrates how using file‑system tools like ls, grep, and write‑file can reduce token waste, enable dynamic storage, improve targeted search, and support continual learning.
This article provides a comprehensive technical guide to Elasticsearch’s core retrieval models—BM25 and TF‑IDF—while detailing modern vector‑based search using dense_vector, kNN, L2 and cosine distances, and demonstrates how to combine keyword and semantic results through hybrid search and Reciprocal Rank Fusion (RRF) with practical configuration examples.
This article presents a comprehensive solution for constructing an enterprise‑level knowledge base using APICLOUD share‑link data, covering data characteristics, system architecture, core algorithms such as streaming token chunking and semantic vector retrieval, performance optimizations, and real‑world integration scenarios.
This article explains what vector stores are, outlines their core workflow of adding, querying, and searching embeddings, compares popular back‑ends like FAISS, Chroma, and Pinecone, and walks through a complete Chinese‑language example using LangChain’s FAISS integration with detailed code and result analysis.
This article explains what embeddings are, how LangChain’s Embeddings interface abstracts various providers, compares common models, and walks through a complete Python example that uses a Chinese‑optimized HuggingFace model to generate document and query vectors, compute cosine similarity, and identify the most relevant text.
This tutorial explains how LangGraph handles short‑term memory with InMemorySaver and thread‑isolated checkpoints, how to persist it using PostgresSaver, and how to build long‑term memory with InMemoryStore and PostgresStore, including semantic search and custom storage extensions.
Memvid replaces traditional vector databases by encoding text chunks as QR codes inside MP4 video frames, achieving up to ten‑fold storage reduction, millisecond‑level semantic search, zero‑infrastructure deployment, and a built‑in conversational interface, while providing a fast‑install Python SDK and CLI.
This article outlines the fundamental RAG workflow—from data indexing and querying to advanced modular designs—highlights key challenges such as retrieval accuracy, model robustness, context limits, and performance, and traces the evolution from naive to modular RAG systems.
This article explains how embedding technology converts unstructured data into vector representations, powers precise retrieval in Retrieval‑Augmented Generation (RAG), outlines the evolution of embedding models, discusses current challenges such as long‑text handling and domain adaptation, and highlights emerging solutions.
This article explains how AI-driven codebase indexing converts massive, undocumented repositories into searchable semantic knowledge bases, detailing the workflow from parsing and embedding to storage and retrieval, and highlighting practical benefits such as faster navigation, code reuse, smarter AI assistants, and historical issue tracing.
This article presents a detailed design and implementation of an LLM‑based intelligent resume matching system that combines semantic vector retrieval, structured rule filtering, multi‑dimensional weighted scoring, and natural‑language interaction to create a fast, quantifiable, and explainable hiring pipeline.
CodeRAG leverages AST‑centric parsing and a hierarchical knowledge graph to overcome text‑only retrieval limits in large code repositories, offering multi‑language analysis, incremental parsing, hybrid indexing, and intelligent context selection for tasks such as code completion, Q&A, documentation generation, and impact analysis.
This article answers practical questions about enterprise semantic search, explaining how Reciprocal Rank Fusion normalizes mixed scoring, how to control vector result size, the trade‑offs of LSH parameters, word‑ and sentence‑based chunking strategies with version‑specific defaults, and flexible embedding dimensionality.
This article walks through the complete design and implementation of a locally deployed, enterprise‑level semantic search system using Ollama for embedding generation and Easysearch for vector retrieval, covering problem analysis, architecture decisions, pipeline configuration, bulk indexing, and hybrid query execution.
This article explores how Ctrip leverages advanced AI techniques—including named entity recognition, entity linking, large language models, and vector search—to replace traditional keyword queries with semantic search, dramatically improving hotel recommendation accuracy and user experience.
The article analyzes the semantic gap in e‑commerce search, compares traditional keyword matching with vector‑based retrieval, and provides a step‑by‑step implementation using Elasticsearch/Easysearch pipelines, embedding models, and a hybrid search strategy to improve user intent understanding.
This article explains why traditional keyword search falls short, introduces Elasticsearch's vector search and hybrid retrieval capabilities, and shows how combining it with large language models enables Retrieval‑Augmented Generation (RAG) for more accurate, context‑aware AI-driven search across text and multimedia data.
This article explains how GraphRAG combines knowledge graphs with retrieval‑augmented generation to overcome the limitations of vector‑only RAG, delivering higher accuracy, better explainability, easier development, and stronger governance for generative AI applications across various domains.
The article reverse‑engineers Cursor’s Agent mode, explains its system prompt, enumerates the 16 built‑in tools, draws insights about the importance of model choice and tool usage, and outlines the components needed to build a similar open‑source coding assistant.
The article explains the mathematical definition of cosine similarity, how large language models use it to compare vector directions, its key characteristics and importance for AI retrieval and generation, and includes a simple Python example using SentenceTransformer.
This article explains why generic text‑embedding models often fail in specialized retrieval tasks, then demonstrates how to fine‑tune such models using contrastive learning, curated job‑listing data, and the Sentence‑Transformers library, achieving near‑perfect accuracy on a job‑matching benchmark.
This article explains how vector embeddings enhance Elasticsearch for semantic search and recommendation, walks through the complete workflow of generating, storing, and querying embeddings, and provides detailed Python and Logstash implementations with code samples, pros and cons, and guidance on choosing the right approach.
This article examines how a front‑end team uses AI innovations—FastGPT visual platform, AI‑powered semantic search, and AI video synthesis—to rebuild business workflows, cut costs, and boost efficiency, highlighting architecture, technical highlights, and practical use cases.
This article explains Retrieval‑Augmented Generation (RAG), its three‑step workflow of retrieval, augmentation, and generation, its key advantages such as improved accuracy and explainability, and compares RAG with traditional pre‑trained models, fine‑tuned models, hybrid models, knowledge‑distillation methods, and RLHF, while also covering vector, full‑text, and hybrid retrieval modes and the role of rerank models.
This article examines the critical role of product reviews for buyers and sellers, outlines the limitations of traditional review processing, and proposes a GenAI‑powered solution—including platform and model choices, batch inference, and semantic search—to efficiently analyze large‑scale e‑commerce feedback.
The article explains how Redis vector search, powered by RedisSearch’s FLAT and HNSW algorithms and supporting various data types and precisions, enables fast AI-driven similarity queries for text, image, and audio, and provides implementation guidance, optimization tips, and a real‑world customer‑service use case.
This tutorial walks through configuring Alibaba Cloud AI services, creating sparse embedding and rerank endpoints, setting up Elasticsearch mappings, indexing Agatha Christie data, and combining semantic search, reranking, and completion APIs to achieve more relevant search results and a RAG‑style answer generation pipeline.
This article introduces DeepSeek, a high‑efficiency search engine optimized for large‑scale data, explains how to configure it within the Cursor database tool using code snippets, and demonstrates its applications such as semantic search, content recommendation, intelligent data analysis, and document similarity matching.
This article explains how combining dense vector‑based semantic search with traditional keyword matching using a unified vector database such as Milvus delivers superior performance, scalability, and simplicity compared to maintaining separate Elasticsearch and vector‑search stacks.
The article analyzes D2LLM’s teacher‑student bi‑encoder architecture and Codefuse‑CGE’s PMA‑enhanced code embedding, showing how both models surpass BERT dual encoders and LLM cross‑encoders in accuracy, efficiency, and storage cost across semantic and code search benchmarks.
This tutorial walks through constructing a zero‑to‑one RAG pipeline that ingests OSS text files via EventBridge, transforms them into embeddings with DashScope, stores vectors in DashVector, and performs semantic search using FunctionCompute and a Qwen‑Turbo LLM, complete with code samples and configuration steps.
Retrieval-Augmented Generation (RAG) combines a retriever that fetches relevant external documents and a generator that uses them, improving LLM accuracy, relevance, privacy, and up-to-date information, but faces challenges such as retrieval latency, computational cost, chunking strategies, embedding selection, and system integration complexity.
This tutorial walks you through configuring Elasticsearch’s Open Inference API to connect with Alibaba Cloud AI Search, covering setup of text generation, rerank, sparse and dense vector services, and demonstrates end‑to‑end requests with code examples for building RAG and semantic search applications.
This article explains how the Volcano Engine DataLeap team leveraged large‑language models to build the “Find Data Assistant”, detailing its design, challenges, embedding‑and‑reranker enhancements, LLM‑driven semantic search, mixing architecture, and practical lessons for improving data asset management and retrieval.
This guide explains vector embeddings, OpenAI's text‑embedding models, and how to use PyMilvus with Zilliz Cloud to create a Retrieval‑Augmented Generation (RAG) system for fast, accurate semantic search.
This article presents a comprehensive overview of 360's document cloud knowledge management and Q&A scenarios, discussing business pain points, large‑model challenges, the advantages of the intelligent document solution, and how knowledge graphs enhance retrieval‑augmented generation and document standardization for AI‑driven enterprise applications.
This article traces Elasticsearch's transformation from a pure text search engine to a versatile hybrid platform that integrates structured, geospatial, aggregation, and vector search capabilities, highlighting its AI-driven innovations, performance optimizations, and growing adoption across enterprises and academia.
From August 11‑16, 2024 the ACL conference in Bangkok featured 14 Ant Group papers covering large‑scale information extraction, decomposed LLMs for semantic search, multimodal hallucination detection, long‑context attention mechanisms, concept‑reasoning datasets, knowledge‑graph alignment, and more, highlighting the group's breadth in natural language processing and AI research.
This article explores how large language models can transform data workflows by using embeddings for semantic search, building private domain knowledge bases, generating SQL code from natural language with visualized results, and enhancing exploratory data analysis, outlining practical steps and benefits for enterprises.
This guide explains how to combine traditional keyword search and vector‑based semantic search into a hybrid retrieval system using Volcengine Cloud Search, covering the underlying concepts, required components, workflow steps, and a complete Python implementation for an image search application.
This article examines how vector search using Milvus handles sentence and paragraph queries compared to traditional Elasticsearch, demonstrating the advantages of embedding‑based semantic matching for document search scenarios through practical experiments and visual results.
This tutorial walks through building a Retrieval‑Augmented Generation application by setting up Baidu’s Vector Database and Qianfan embedding service, configuring credentials, creating a document database and vector table, loading and chunking PDFs, generating embeddings, storing them, and performing scalar, vector and hybrid similarity searches, ready for integration with Wenxin LLM for answer generation.
This article explains how traditional keyword search with Elasticsearch often yields inaccurate or incomplete results for document retrieval, introduces vectorization and semantic search using NLP embeddings, and demonstrates a practical workflow that combines these techniques with the Milvus vector database to achieve more accurate and efficient document search.
This article reviews the author’s hands‑on experience with Pinecone’s serverless vector database, various embedding and generation models such as all‑MiniLM‑L6‑v2, text‑embedding‑ada‑002, clip‑ViT‑B‑32, and GPT‑3.5‑turbo‑instruct, and demonstrates how they are applied to semantic search, RAG, recommendation, hybrid, and facial similarity tasks using Python code examples.
Retrieval‑augmented generation (RAG) enhances large language models by fetching up‑to‑date, authoritative information from external sources, addressing hallucinations, outdated knowledge, and lack of citations, while offering cost‑effective implementation, improved relevance, user trust, and greater developer control through vector databases, semantic search, and prompt engineering.
This article explores the emerging enterprise knowledge assistant paradigm in the era of large models, detailing traditional knowledge management challenges, solution architecture using vector databases and LLMs, core technologies such as ETL pipelines, reranking, secure fine‑tuning, and future prospects for intelligent enterprise applications.
Vector databases gain traction because they dramatically reduce storage, learning, scaling, and large‑model limitations costs by enabling semantic similarity search, RAG‑based prompt optimization, efficient high‑dimensional indexing, and cloud‑native architectures, making them essential for modern AI applications despite the promotional context.
This article explores how large language models can boost data workflows by using embeddings for semantic retrieval, building domain‑specific knowledge bases for private Q&A, generating SQL code from natural language, and automating exploratory data analysis, offering practical steps and visual examples.
This article explains the design and implementation of the Codebase AI Assistant in the Chocolate Factory framework, covering its problem‑solving DSL, retrieval‑augmented generation pipeline, indexing and querying stages, prompt strategies, and code‑splitting rules that together enable efficient semantic code search.
The article explains how combining vector databases with large language models can help governments and enterprises leverage massive private data for AI, covering semantic search, approximate nearest neighbor techniques, alignment challenges across modalities, and future directions for fine‑grained data integration.
This article explores the rapid evolution of Baidu's semantic search models, the large GPU consumption they entail, and how extensive INT8 quantization, sensitivity analysis, calibration data augmentation, hyper‑parameter auto‑tuning, and advanced methods like Quantization‑Aware Training and SmoothQuant dramatically improve inference performance while preserving business metrics.
Baidu applied large‑scale INT8 quantization to its ERNIE search semantic models, achieving over 25% inference speedup with less than 1% degradation in relevance metrics by selectively quantizing less‑sensitive fully‑connected layers, using automated calibration, hyper‑parameter tuning, and techniques such as QAT and SmoothQuant, while paving the way for even lower‑bit quantization and token pruning.
This article explains what embeddings and vector databases are, why they matter for large language models, and how they can be used to overcome token limits by storing and retrieving relevant text chunks for accurate, context‑aware responses.
This article explains what embeddings and vector databases are, how they are generated with models like OpenAI's Ada, why they enable semantic search and help overcome large language model token limits, and demonstrates a practical workflow for retrieving relevant document chunks using cosine similarity.
This presentation explains where knowledge resides in data architectures, demonstrates knowledge‑graph‑driven skill discovery, metadata management, and semantic search, and concludes with a comparison of GraphQL and Cypher for graph queries, illustrated with real‑world Neo4j case studies.
This article introduces VentureSights, an AI‑driven intelligent industry analysis platform built on knowledge‑graph technology and the concept of industry atoms, detailing its core modules, workflow, industry‑atom representation, extraction algorithms, and overall system architecture for generating comprehensive industry reports and insights.
This article explains Meituan's 'Meituan Brain' initiative, detailing the construction of life‑service knowledge graphs—including tag and dish graphs—through data mining, semantic extraction, synonym discovery, graph labeling, and applications such as open QA, search ranking, and recommendation using AI and GNN techniques.
This article outlines the development of a power‑sector knowledge graph by China Electric Power Research Institute, covering the AI institute overview, the background of power knowledge engineering, methods for representation and ontology construction, practical applications in dispatch, inspection and customer service, and future challenges.
This article presents JD's end‑to‑end semantic search recall pipeline, covering multi‑stage recall, a dual‑tower embedding model with multi‑head attention, a heterogeneous graph neural network (SearchGCN), a transformer‑based synonym generation system, and a joint index‑training approach that integrates product quantization to improve recall accuracy and efficiency.
The talk presents JD's end‑to‑end semantic search recall pipeline, covering multi‑stage retrieval, a dual‑tower embedding model with multi‑head attention, a heterogeneous graph neural network for low‑frequency items, automatic synonym generation via transformer models, and a joint training approach that integrates product quantization directly into the model to improve accuracy and efficiency.
Xianyu search recall stage upgraded from simple text matching to semantic vector retrieval using DSSM with Electra‑Small, query graph attention, and behavior‑based I2I, adding structured attributes and OCR tags, improving AUC to 0.824 and HitRate@10 to 90.1%, boosting purchase metrics by up to 4%.
The article outlines five major trends—neural-network-enhanced search, semantic search, document understanding, image and voice search, and knowledge graphs—that are transforming enterprise use of unstructured data by leveraging AI to deliver precise, context-aware answers and insights.
This article walks through building a large‑scale semantic search system on the English Wikipedia using the Weaviate vector database, GraphQL queries, and pre‑trained Sentence‑BERT and BERT Q&A models, covering dataset preparation, schema design, import pipelines, query examples, and production deployment strategies.
This article provides a comprehensive overview of knowledge graphs and entity linking, covering their definitions, practical uses in question answering, search and recommendation, the standard pipeline of mention detection, candidate generation and scoring, challenges such as scalability and multilinguality, and recent research advances including dual‑encoder, RELIC, deep retrieval, and multilingual BERT‑based models, followed by a discussion of modern knowledge‑graph construction methods.
WanFang upgraded its academic paper retrieval system by adopting PaddleNLP’s Chinese pre‑trained Sentence‑BERT models, using weakly supervised SimCSE data and Milvus vector indexing, compressing the transformer for TensorRT‑accelerated inference, achieving 70% better matching quality and 2600 QPS latency‑optimized performance.
The article explains why traditional text‑based search engines like ElasticSearch struggle with modern multimodal data, introduces vector databases that store implicit semantic embeddings, and proposes a generalized search architecture that decouples data‑to‑vector mapping from the engine while leveraging clustering or graph indexes for similarity search.
This article details Zhihu Search's engineering solutions for long‑tail query challenges, covering historical development, term weighting, synonym expansion, query rewriting with reinforcement learning, and semantic recall using BERT‑based models, while also outlining future research directions such as GAN‑based rewriting and lightweight pre‑training.
This article presents a comprehensive overview of 58 Search's multi‑stage retrieval system, compares term‑match and semantic matching, details the design, training, and optimization of interactive, dual‑tower, and semi‑interactive BERT‑based semantic models, and discusses their practical deployment in ranking and recall stages.
This article explains how financial institutions can leverage large amounts of structured and unstructured data to build and apply financial knowledge graphs, covering AI key technologies, schema design, data extraction, graph construction, storage solutions, and real-world use cases such as intelligent tagging, recommendation, policy analysis, and executive relationship mining.
