Knowledge Graph Application Cases: Meituan Brain, Sage Knowledge Base, and Other Industry Examples

Knowledge graphs are special graph structures that combine semantic information with graph topology, and they have been widely adopted in recommendation, healthcare, and many other industry scenarios.

Case 01 – Meituan Brain (Lifestyle Service Knowledge Graph) – Since 2018 Meituan’s NLP Center has built a massive knowledge graph for the lifestyle and entertainment domain to improve merchant and user experience. The main challenges are strong explainability requirements, diverse and sparse data, and complex spatio‑temporal contexts. The solution framework includes:

Graph‑structured information display : e.g., extracting drug effects from prescriptions or presenting flavor‑based snack filters.

Graph‑path‑guided recommendation : (a) direct graph‑path recall, recommending lower‑level entities linked to the query entity (e.g., “milk tea” → “pearl milk tea”, “vanilla milk tea”); (b) embedding‑based recall, constructing a graph from user query history and clicked POIs, training a GNN to obtain embeddings, and retrieving nearest‑neighbor query vectors.

Knowledge‑based reasoning for recommendation reasons : using the graph to infer user‑merchant paths and generate explanations such as “People from Sichuan like this restaurant’s boiled fish”.

Case 02 – Fourth Paradigm Sage Knowledge Base – A low‑threshold, end‑to‑end knowledge‑graph platform that supports applications such as QA, recommendation, drug discovery, and stock prediction. It introduces automated knowledge‑graph representation learning from triples to subgraphs.

Model design highlights:

Triple‑based models : translation‑based (e.g., TransE), MLP‑based, ConvE, RSN, bilinear models, and AutoSF/AutoSF+ which automatically search for optimal relation matrices using progressive or genetic‑algorithm‑based search.

Relation‑path models : PTransE extends TransE by modeling multi‑relation paths; RSN uses RNN with skip connections to capture long‑term dependencies; Interstellar splits paths into triples and searches optimal sub‑models.

Graph Neural Network models : R‑GCN, CompGCN, KE‑GCN (shallow embeddings + relational GNN), GraIL (inductive subgraph reasoning without pretrained embeddings), and RED‑GNN (enhanced relation subgraphs with dynamic‑programming‑based parallel computation).

Case 03 – Other Knowledge‑Graph Applications – Examples include logic‑rule reasoning (RNNLogic) from Mila AI Lab, and a power‑industry knowledge graph from China Electric Power Research Institute.

For more case studies, scan the QR code below to download the full "AI Technology Application Cases" manual.

Thank you for reading; please like, share, and follow for more AI technology insights.