This article presents a study on enhancing text representation and clustering in a small‑sample setting for 58's second‑hand car intelligent customer service by introducing a Bi‑LSTM based pre‑training language model and an improved Deep Embedded Clustering (DEC) algorithm, demonstrating significant gains in accuracy, silhouette score, and answer‑rate through extensive experiments.
The article recounts an Alibaba tech employee's typical workday, covering paper reading, massive data handling, model building, corporate knowledge absorption, and offers a detailed NLP learning roadmap with practical advice for students and newcomers to the field.
This article describes the design, model selection, experimental evaluation, and engineering optimization of an AI‑driven post‑sale question‑answering system for hotel services, covering FAQ construction, intent detection, deep‑learning matching models such as DSSM, ESIM, BERT, and their performance and latency trade‑offs.
When you browse video apps by tags, receive personalized recommendations, or search with keywords, the seamless experience is powered by Natural Language Processing, which analyzes and interprets textual data to connect users with relevant content, and the article invites you to scan a QR code for further exploration.
Query understanding (QU) involves lexical, syntactic, and semantic analysis of user queries to enable effective search recall and ranking, covering modules such as preprocessing, correction, expansion, segmentation, intent detection, term importance, and guidance, with detailed discussion of algorithms, models, and system architecture.
This article presents a comprehensive overview of automated techniques for generating high‑quality search engine advertising keywords, covering background, traditional manual methods, intelligent keyword expansion using NLP, segmentation, POS tagging, BILSTM‑CRF, BERT classification, semantic matching with DSSM, and additional approaches such as query suggestion and synonym rewriting.
This article outlines an AI-powered workflow for constructing and enriching a business knowledge base in intelligent customer service, covering preprocessing, intent detection, deep and shallow semantic feature engineering, hierarchical bucket clustering, and automated summary extraction to improve FAQ coverage and reduce manual workload.
This article reviews recent research on machine translation that tackles the training‑inference distribution gap, exposure bias, and slow autoregressive decoding by introducing scheduled sampling, differentiable sequence‑level losses, cube‑pruning, and sequence‑aware non‑autoregressive decoding, showing BLEU gains and significant speedups.
In this interview, Didi AI Labs’ Han Kun explains how the DELTA platform unifies TensorFlow‑based NLP and speech models—supporting tasks from text classification to voice emotion recognition—through a modular, easily deployable architecture, accelerating development, powering Didi products, and now open‑sourced for broader AI collaboration.
The article surveys iQIYI’s multimodal short‑video content‑tagging pipeline, detailing extraction‑ and generation‑based methods, challenges of open‑world tags, model evolution from rule‑based to Transformer generators, visual‑text fusion techniques, and applications such as recommendation, search, clustering, and future enhancements.
This article describes how Gaode Map leverages machine‑learning techniques—such as word2vec embeddings, LSTM networks, fine‑tuning, and confidence‑threshold ensembles—to automatically classify and verify massive user‑feedback intelligence, streamlining the multi‑step workflow from data collection to road‑map updates and dramatically improving efficiency.
This article explores how integrating unstructured business data into structured knowledge graphs and applying graph neural networks can overcome deep‑learning bottlenecks in NLP, dramatically improving HS‑code product classification accuracy from around 60% to over 75% through richer reasoning and multimodal knowledge.
This article provides a comprehensive QA‑style review of entity‑relation extraction (ERE), covering pipeline drawbacks, various decoding strategies for NER, common relation‑classification techniques, shared‑parameter and joint‑decoding models, recent transformer‑based approaches, challenges such as overlapping entities, low‑resource settings, and the use of graph neural networks.
This article systematically reviews recent developments in text data augmentation for natural language processing, covering common scenarios such as low‑resource and imbalanced classification, and detailing five major techniques—including back‑translation, EDA, TF‑IDF‑based replacement, contextual augmentation, and language‑model‑based methods—with experimental results and future directions.
This article proposes a method to identify low-quality text in business data by training a large-scale unsupervised language model to compute sentence perplexity, converting the detection problem into a threshold decision, and details model design, challenges, optimizations, and online performance results.
This article presents a comprehensive overview of tag mining practices—including similarity‑based, compound‑word, topic, hot‑search, and image‑based approaches—along with detailed evaluations of Chinese segmentation tools and systematic tag optimization techniques such as synonym and negative‑word detection.
This presentation reviews adversarial training techniques for transformer‑based NLP models, covering the motivation, image‑based and text‑based attack generation, standard PGD, its variants FreeAT and YOPO, the proposed FreeLB method, extensive GLUE experiments, and conclusions about robustness and future directions.
This article explains how Gaode Map’s search engine combines traditional NLP techniques with map‑specific intent parsing, machine‑learning‑driven recall and ranking, and continuous model upgrades to dramatically improve POI retrieval accuracy across location, search and navigation scenarios.
The article explains that natural language understanding is fundamentally a structure‑prediction problem whose difficulty stems from language innovation, recursion, ambiguity, subjectivity and social factors, and argues that richer semantic representation spaces and multimodal context modeling are needed for true AI comprehension.
The article explains how Tencent tackles content understanding in feed recommendation by evolving from traditional classification, keyword, and entity methods to a multi‑layer interest graph that captures concepts and events, addressing the need for full context, reasoning about user intent, and improving online performance.
The article details Ctrip’s Marco Polo content platform, describing its data, algorithm, and functional layers, and explains how AI techniques such as NLP, semantic matching, named‑entity recognition, and image classification are applied to automate product‑centric content mining, article generation, quality rating, and first‑image selection.
Amap replaced its rule‑based feedback pipeline with a three‑stage, LSTM‑driven system that combines word2vec embeddings and structured fields, achieving over 96% classification accuracy, cutting manual workload by 80%, and slashing per‑task costs while enabling scalable, data‑driven map quality improvements.
This article details Zhihu's search engineering advances over the past year, covering long‑tail query challenges, term‑weight calculation, synonym expansion, query rewriting with translation models and reinforcement learning, and semantic retrieval using BERT‑based embeddings, while outlining future research directions.
This article reviews the specific NLP difficulties encountered in Sina Weibo's feed—such as short text, informal language, and ambiguous user behavior—and details the multi‑stage tagging system, material library, multimodal modeling, multi‑task learning, and large‑scale pre‑training techniques used to address them.
The article discusses how to discover hot topics using algorithms such as TextRank, BERT embeddings, and BM25, outlines the lifecycle of a hot topic, and proposes practical methods for evaluating article deduplication accuracy and recall when no labeled test set is available.
This article reviews the evolution of language models from bag‑of‑words to BERT, compares ELMo, GPT, and BERT architectures, and details how NetEase Yanxuan applies pre‑trained models to classification, text matching, sequence labeling, and generative tasks in production.
This article explains how Tencent tackles content understanding for personalized feed recommendation by combining traditional classification, keyword, and entity methods with deep learning embeddings, introducing an interest graph composed of taxonomy, concept, entity, and event layers to capture full context and infer user consumption intent.
The article highlights several ingenious pre‑deep‑learning techniques for NLP, including the Distributional Hypothesis, Bag‑of‑Words, Latent Semantic Analysis, Probabilistic Topic Models, BMES/BIO tagging schemes, and TextRank, explaining their principles, advantages, and historical significance in text representation and processing.
The article details Amap’s upgraded map‑search query analysis, introducing a two‑stage city‑identification system, enhanced where‑what segmentation with CRF and GBDT models, a three‑stage path‑planning pipeline, and outlines future deep‑learning and knowledge‑graph enhancements for robustness and low‑frequency query handling.
TurboSearch is Tencent AI Lab's next-generation large-scale search system, delivering distributed massive indexing, high-performance parallel retrieval, multi-granularity and multi-modal vector indexing, private Docker deployment, integrated NLP query analysis, extensible plugins, and robust operations for massive data and diverse search scenarios.
This article describes the end‑to‑end development of a multi‑class NLP system for Ctrip airline ticket customer service, covering problem analysis, data preprocessing, sample balancing, model architecture (TextCNN and Bi‑GRU), training strategies, performance evaluation, and online customization to achieve high accuracy in intent recognition.
Recent advances in map‑search text processing replace rule‑based pipelines with machine‑learning and deep‑learning models for query analysis, error correction, rewriting, and omission, using phonetic and spatial entity correction, vector‑based similarity, and CRF sequence labeling within a three‑stage architecture of analysis, recall, and ranking to deliver more precise POI results.
This article explains how query understanding (QP) transforms user search queries into structured semantic blocks and intent categories using rule‑based NLP, entity recognition, and post‑processing, and describes its taxonomy, implementation details, and practical impact on search engine results.
This article describes the development of MT‑BERT, a BERT‑based language model pre‑trained on Meituan‑Dianping business data, its distributed mixed‑precision training pipeline, domain adaptation, knowledge‑graph integration, model compression techniques, and the wide range of downstream NLP applications achieved in the platform.
MT‑BERT at Meituan‑Dianping combines mixed‑precision, domain‑adapted continual pre‑training, knowledge‑graph‑aware masking, and extensive compression techniques to produce fast, accurate BERT models that power fine‑grained sentiment analysis, intent classification, recommendation reasoning, and other NLP tasks across the platform.
This article introduces the Huajiao platform's text anti‑spam architecture, analyzes spam categories and challenges, compares rule‑based and machine‑learning approaches, details traditional NLP methods and the TextCNN deep‑learning model, provides its TensorFlow implementation, and describes the online deployment workflow.
This article examines how sequence labeling methods such as pattern matching, CRF, and deep‑learning models like BiLSTM‑CRF and BERT are applied to automotive search tasks—including car‑series, model, and location/entity recognition—detailing their development, implementation challenges, and performance results.
In an interview, Gitee’s chief architect Chen Xin explains why Elasticsearch was chosen for code search, outlines how combining search with NLP can both aid semantic understanding and enrich repository queries, and shares his views on the platform’s fast‑evolving ecosystem and upcoming community meetup.
This article introduces GeoBERT, a novel pre‑training method for Chinese address strings that leverages seven jointly constrained tasks to capture spatial semantics, administrative hierarchy, and similarity relationships, enabling downstream address classification, segmentation, POI extraction, similarity comparison, and authenticity verification with reduced annotation dependence.
This article presents a comprehensive solution to a competition that requires matching description paragraphs with the three most relevant papers from a 200,000‑paper corpus, detailing background, task definition, evaluation metrics, modeling strategy, and core algorithms such as SIF, InferSent, Bi‑LSTM, and BERT.
Using Python to scrape 150,000 NetEase Cloud Music folk lyrics, the author applies jieba segmentation, word‑frequency analysis, word‑cloud visualization, seasonal and geographic preferences, and SnowNLP sentiment scoring to reveal the themes, hopes, and city loves of Chinese folk singers.
This talk analyzes the technical and practical challenges of current chatbots, introduces the next‑generation paradigm of virtual life, and details how AI technologies such as speech, vision, NLP, and knowledge graphs can be combined with multimodal interaction to create more intelligent and engaging conversational agents.
This article details Ctrip's AI‑driven Marco Polo platform, describing how large‑scale NLP pipelines combine extraction, richness evaluation, semantic matching and deep‑learning generation (CopyNet, TA‑seq2seq) to produce high‑quality recommendation reasons across multiple product scenarios.
This article presents a comprehensive overview of the challenges, requirements, and technical solutions for multilingual content understanding in UC's international information‑flow recommendation system, covering structured signal construction, low‑resource NLP techniques, transfer learning, quality modeling, and image‑based signal integration.
This article presents a comprehensive overview of Alibaba's brand advertising business model, its technical architecture, and the practical application of NLP algorithms—specifically brand intent recognition and short‑text relevance—detailing model evolution, evaluation results, and future research directions.
This article systematically reviews the history, core concepts, variants, and practical implementations of attention mechanisms—from early additive and multiplicative forms to self‑attention, multi‑head attention, and recent transformer‑based models—highlighting why attention has become fundamental in modern AI research.
This article explains the motivation, challenges, architecture, and algorithms behind building a large‑scale cognitive concept graph for search, detailing data construction, concept mining, fusion, confidence scoring, hierarchical structuring, validation, service algorithms, platform access, and real‑world applications such as intent recognition and entity recommendation.
Xianyu tackled slow seller responses, inconsistent answers, and bargaining friction by building an AI-driven auto-reply system that extracts product attributes, generates knowledge bases, and uses intent recognition and response generation models, resulting in faster replies, higher conversion rates, and reduced seller workload.
This article explains the design and evaluation of RE2, a lightweight yet expressive text‑matching framework for FAQ‑style chatbots, detailing its five‑layer architecture, block‑wise residual connections, experimental results on SNLI, MultiNLI, SciTail, Quora and WikiQA datasets, and its significant performance improvements in Alibaba’s DingXiaoMi service.
The talk details how Youku’s Beidou Star AI platform deconstructs the drama “Chang’an Twelve Hours” with NLP, computer‑vision, knowledge graphs and multi‑task deep models to quantify script, character and emotion uncertainty, enabling predictive scoring that lifted the series’ daily index above one million and outlines future hybrid decision‑engine research.
This article explains how to create a functional spell‑checking tool using a 21‑line Python script that leverages Bayesian probability and a large word‑frequency corpus to suggest correct spellings for misspelled inputs.
This article provides a comprehensive, beginner‑friendly overview of attention mechanisms—from their origins in early neural machine translation papers to modern self‑attention, multi‑head attention, and transformer variants—explaining core concepts, common variants, and why attention has become essential across NLP and vision tasks.
On August 6, Dr. Feng Ao presented a comprehensive overview of the evolution of word representations and sentiment analysis, illustrating the shift from traditional linguistic features to modern pretrained models such as BERT and XLNet, and sharing practical convolutional experiments relevant to industry applications.
At ACL 2019, Didi unveiled DELTA, an open‑source TensorFlow‑based training framework that unifies NLP and speech tasks, offers configurable pipelines, benchmark models, and seamless deployment, enabling AI developers to quickly move from research to production while leveraging Didi’s extensive open‑source ecosystem.
This article presents a comprehensive solution to the absolute semantic recognition competition, detailing the problem background, dataset, evaluation metrics, feature engineering, model architecture—including Attention, Capsule, Bi‑GRU, and BERT—and analysis of results and lessons learned.
The JDATA Absolute Semantic Recognition Challenge concluded on July 15 with eight finalist teams presenting their models, the champion "Label Team" using BERT, judges highlighting data cleaning, model fusion, and the distinction between absolute and contextual words, and a detailed ranking of all participating groups.
This article systematically reviews the evolution of attention mechanisms—from early additive and multiplicative forms to self‑attention and multi‑head variants—explaining their core three‑step framework, key differences, and why they have become essential across NLP, vision, and broader AI applications.
The article presents TaskBot, a modular task‑oriented dialogue robot that uses a Bi‑LSTM‑CRF joint intent‑entity model and a state‑machine based dialogue manager to handle multi‑turn conversations such as flight booking or rental housing, detailing its architecture, implementation, and performance.
This article explains how Beike, China’s largest residential service platform, leverages its massive house, client, and text data to design a multi‑layered tag architecture, detailing data sources, tag construction methods—including classification, keyword, geographic, anonymous topic, and temporal tags—and their application to improve personalized house search and recommendation.
This article examines XLNet, contrasting it with BERT by detailing its novel permutation language modeling, dual‑stream attention, and larger pre‑training data, and analyzes experimental results that show XLNet’s superior performance on reading‑comprehension, GLUE, and other NLP tasks, especially for long documents.
This article reviews the evolution of word and sentence embeddings, covering foundational theories like vector semantics and distributional hypothesis, practical models such as Word2Vec, GloVe, fastText, Skip‑Thought, Quick‑Thought, and evaluation techniques, while offering implementation tips and real‑world use cases.
This article surveys the rapid proliferation of BERT-based research over the past six months, analyzing its impact on various NLP tasks such as question answering, information retrieval, dialog systems, summarization, data augmentation, classification, and sequence labeling, while also discussing the model's strengths, limitations, and future research opportunities.
This article reviews the historical development of natural language processing language models, covering expert rule‑based systems, statistical n‑grams, smoothing techniques, neural network models such as NNLM, RNN, word2vec, GloVe, ELMo, and the transformer‑based breakthroughs of GPT, BERT and GPT‑2, and summarizes their impact on modern NLP tasks.
This article reviews modern deep‑learning methods for Chinese word segmentation, comparing traditional CRF‑based approaches with BiLSTM‑CRF and BERT models, describing their architectures, training procedures, experimental results, and practical considerations for deployment.
This article reviews the evolution, architecture, and key technical challenges of general‑domain conversational QA systems, describing Alibaba UC’s search background, dialogue bot types, data pipelines, and advanced methods such as transfer learning, few‑shot learning, and multi‑dimensional dialogue management.
This article reviews the taxonomy of transfer learning in NLP, summarizes representative pre‑training models such as ELMo, ULMFiT, BERT, GPT, MASS and UNILM, discusses their strengths and limitations, and provides practical recommendations for applying these techniques in real‑world projects.
This article walks through the development of deep learning for natural language processing, starting with basic neural cells and shallow networks, then exploring CNNs, RNNs, LSTMs, TextCNN, ESIM, ELMo, and culminating with the Transformer‑based BERT model, its training objectives, fine‑tuning strategies, and performance comparisons.
This article introduces Grammatical Error Correction (GEC), explains its main error categories, outlines four key challenges, reviews evaluation metrics and the evolution of NLP approaches, and showcases practical solutions and product applications developed by Liulishuo.
At QCon Beijing 2019, four Beike technology experts presented the practical use and challenges of machine learning for user profiling, deep‑learning‑based house‑quality scoring, intelligent customer‑service systems, and AI‑driven floor‑plan generation, summarizing the architecture, data pipelines, model evolution, and future improvement directions.
The article details the Travel team's award‑winning solution for the WSDM Cup 2019 fake‑news detection task, describing data analysis, preprocessing, label‑propagation augmentation, a BERT‑based baseline, a three‑stage multi‑level model‑fusion framework, experimental results, and future directions.
The talk introduces Alibaba’s Fish Brain platform, an AI‑powered decision‑support system for entertainment that combines a three‑layer data‑model, AI‑processed basic data, and application models, leveraging NLP, computer‑vision, custom embeddings, loss functions and predictive hybrid networks to analyze content, user behavior, and forecast performance.
The AI Scientific Frontier Conference 2019, co‑hosted by the Chinese Academy of Sciences AI Alliance and Beijing Institute of Technology, gathers leading researchers to present cutting‑edge talks on AI theory, deep learning, computer vision, robotics, NLP, big data, and related applications, with detailed schedules, speaker bios, venue information, and registration details provided.
Natural language processing, a cornerstone of AI, relies on language models to capture linguistic features; this article reviews classic pre‑training models—ELMo, GPT, and BERT—explaining their architectures, training objectives, and how they boost downstream NLP tasks despite data‑scarcity challenges.
This article demonstrates how to extract the character relationships from the classic novel Romance of the Three Kingdoms using Python, building entity dictionaries, constructing a co‑occurrence social network with HarvestText and NetworkX, visualizing subgraphs, ranking important figures, detecting communities, and creating an animated view of the network’s evolution over the story.
The article explores how chatbots require personalized dense knowledge graphs, dynamic temporal graphs, subjective emotion modeling, integration with external services, and multimodal media support, while also promoting a new NLP book and a related giveaway for readers.
Meituan’s information flow content team leverages advanced text‑generation techniques—including extractive and generative models, context‑aware representations, and Seq2Seq with attention—to overcome creative constraints, improve evaluation, and boost business metrics such as click‑through rates and user engagement through optimized titles and merchant copywriting.
This article reviews the history of natural language processing, explains key deep‑learning models such as NNLM, Word2vec, CNN, RNN, attention mechanisms, and Transformers, and discusses their applications, future trends, and practical considerations in NLP tasks.
This article explains a flexible template‑matching approach that dynamically extracts concise, user‑friendly tags from online travel reviews, detailing the system architecture, key concepts, step‑by‑step implementation, and matching rules that improve recall and relevance.
This article provides an in‑depth overview of BERT, covering its NLP background, GLUE benchmark achievements, Transformer‑based architecture, pre‑training strategies (MLM and NSP), downstream fine‑tuning methods, and includes detailed PyTorch code implementations of its core components.
This article recounts a Hulu NLP research engineer's experience, reviewing key milestones such as NNLM, Word2vec, Transformer and BERT, and then contrasting academic research with product development while illustrating real-world projects like news personalization and content embedding, and describing the supporting AI platform architecture.
Vipshop's in‑house sentiment monitoring platform integrates web‑scraped reviews, WeChat comments and internal service messages, applying lexical sentiment scoring, dictionary‑based Chinese word segmentation, TF‑IDF keyword ranking and lightweight classification to deliver real‑time insights, alerts and actionable reports for thousands of daily user comments.
In the WSDM Cup 2019 fake-news detection challenge, the Meituan Travel team secured second place by combining extensive data analysis, Chinese-English BERT fine-tuning, label-propagation augmentation, and a three-level fusion framework—blending, stacking, and linear regression—that lifted weighted accuracy to 0.88156.
Dr. Wang Zhongyuan, a Renmin University graduate turned AI visionary, built a world‑class NLP center and a 1.8‑billion‑entity knowledge graph at Meituan after pioneering knowledge‑graph research at Microsoft Research Asia and leading Facebook’s entity‑linking service, emphasizing solid fundamentals, data integration, and patient, continuous learning for AI impact.
Cheng Huige’s winning AI Challenger 2018 solution treated fine‑grained Chinese review sentiment as a 20‑aspect multi‑class task, combining a high‑capacity LSTM encoder with self‑attention, word‑and‑character embeddings, simplified ELMo pre‑training, diverse tokenizations and a weighted seven‑model ensemble (including BERT), which together delivered the competition’s top F1 performance.
This article explains BERT’s architecture—from its bidirectional Transformer encoder and attention mechanisms to its pre‑training tasks—and presents extensive experiments showing its superior performance on various Chinese and English text‑classification benchmarks across multiple datasets.
The article explains why Chinese word segmentation is essential for NLP tasks, outlines its fundamental difficulties such as ambiguity and out‑of‑vocabulary words, reviews dictionary‑based, statistical, and CRF approaches, and shares practical experiences from 58 Search’s production system.
The article explains how Tmall’s new “Most Concerned” module leverages NLP techniques, fastText embeddings, Bi‑LSTM classifiers, and a custom clustering algorithm to filter, group, and link consumer questions with relevant product reviews, improving the shopping experience across many product categories.
The talk explains how Meituan transformed its O2O hotel search by layering a multi‑stage retrieval pipeline with intent‑aware NLP, then progressively upgrading ranking—from XGBoost to MLPs, feature‑embedding networks, and finally a Wide‑Deep multi‑task model—while tackling data sparsity, diverse scenarios, and deploying the system via TensorFlow‑Serving and the in‑house MLX platform.
iQIYI’s NLP pipeline—combining CRF‑based segmentation, bidirectional LSTM/GRU models with attention and a CNN classifier—automatically extracts opinion targets, sentiment words and polarity from unstructured video comments, aggregates them across users to reveal collective attitudes toward actors, plot, and visual effects, and guides future work on implicit opinions and broader sentiment domains.
This article explains what a tourism knowledge graph is, discusses its architecture, construction methods, practical applications such as QA and recommendation, and explores future directions integrating knowledge graphs with deep learning and multi‑domain fusion.
This article details the business background, technical architecture, and practical deployments of natural language processing for content mining at Yuewen, covering tag construction, knowledge‑graph building, role analysis, recommendation generation, porn and plagiarism detection, and summarizing lessons learned.
Tencent Xiaowei is an open, easy‑to‑integrate conversational AI platform that combines NLU, dialogue management and generation, supports multi‑turn context via Memory Networks, uses bidirectional RNN and CNN‑based intent classifiers, and powers smart speakers, TVs and customer‑service bots by leveraging Tencent’s rich content ecosystem.
This article describes the challenges and solutions encountered while developing an AI-driven medical dialogue assistant, covering data acquisition, preprocessing, model selection such as DCNN and Bi‑LSTM‑CRF, question generation, and system architecture, with insights applicable to similar healthcare chatbot projects.
This article presents an overview of machine reading comprehension, covering the evolution from modular question‑answering systems to end‑to‑end neural models, discusses key datasets such as SQuAD and MS MARCO, and explores voice interaction technologies and future industry trends.
An in‑depth overview of multi‑task learning for natural language processing is presented, covering deep learning foundations, challenges, various multi‑task learning paradigms (hard, soft, shared‑private, function‑level, hierarchical, and search‑based sharing), benchmark platforms, and future research directions, illustrated with numerous diagrams.
The talk outlines how Beike built a real‑estate conversational AI platform, covering the market need for dialogue systems, the five technical challenges, data‑driven intent and slot extraction, model choices such as FastText and Bi‑LSTM‑CRF, a three‑layer system architecture, multi‑intent handling, and future directions like 4D viewing and an internal AI dialogue platform.
Meituan Brain builds a massive multi‑modal knowledge graph of billions of entities and triples across food, entertainment, and travel, using advanced extraction, validation, fusion, and reasoning techniques to empower search, recommendation, merchant tools, and fraud detection while addressing scalability and schema‑evolution challenges.
The third iQIYI Tech Salon in Beijing, held despite strong winds, showcased five expert talks on AI‑driven video library management, NLP for entertainment content, AI‑based video encoding, traffic anti‑fraud systems, and short‑video personalized recommendation, illustrating AI’s impact on content creation, quality, security, and user experience.
Meituan’s AI platform leverages a massive “Meituan Brain” knowledge graph—combining a common‑sense graph and an encyclopedia‑scale graph of billions of reviews, millions of dishes and merchant tags—integrated with deep‑learning models to enable interpretable, cross‑scenario search, recommendation, sentiment analysis and AI‑driven dining assistance.
The article presents a comprehensive overview of Meituan's intelligent customer service system, detailing its evolution, the roles of QABot, TaskBot and ChatBot, the underlying NLP and knowledge‑graph technologies, model implementations such as DSSM and Seq2seq, and the impressive operational results achieved in food‑delivery and ride‑hailing services.