Construction and Application of Meituan's Life Service Knowledge Graph

Meituan, China’s largest online local life service platform, has been developing a comprehensive knowledge graph called "Meituan Brain" since 2018 to enhance user experience across dining, delivery, hotel, and other domains.

The roadmap began with restaurant knowledge graph construction, progressed to tag graphs for extracting user intent from reviews, and later expanded to dish and cross‑domain graphs, now covering billions of entities and triples.

Tag Knowledge Graph : Uses simple sequence labeling for tag extraction, combines semantic and contextual discrimination, and applies distant supervision with voting to obtain precise tags from massive user comments. The pipeline includes knowledge extraction, relation mining, graph labeling, and application.

Relation mining focuses on synonym discovery by vector indexing, hashing recall, and a synonym discrimination model, achieving high accuracy and efficiency.

Graph labeling connects tags with merchants using frequency thresholds and a merchant‑labeling module to filter noisy associations, employing BERT‑based models that consider tag, merchant, comments, and taxonomy information.

Applications include open‑domain Q&A, search recall and ranking (by mapping user queries to tag IDs), and representation‑based enhancements using GraphSAGE GNNs that model Query‑POI and Tag‑POI edges, yielding significant online performance gains.

Dish Knowledge Graph construction targets systematic dish understanding through name parsing, deep‑learning inference, explicit reasoning, and multimodal pre‑training. Techniques involve span‑based token labeling, BERT/ResNet encoders, contrastive learning, and semi‑supervised multimodal models to predict dish attributes and improve SKU initialization.

Overall, the knowledge graph framework enables richer semantic connections between users, merchants, dishes, and contexts, supporting various downstream tasks such as recommendation, search, and conversational AI within Meituan’s ecosystem.