TCM LLM · Patents
TCM LLM patents
Invention patents on TCM LLMs, knowledge graphs, RAG, intelligent inquiry, and prescription recommendation—not a dump of herbal-formula patents.
TCM dialogue generation with a large language model and long-term memory
Generates personalized TCM dialogue from a long-term memory bank and user portrait (published, not granted).
Digital inheritance and analysis of Zhang Xichun's theory, method, formula, and herbs
Rule engine plus three-way retrieval and an LLM for Zhang Xichun school analysis (published, not granted).
Causal contrastive learning and LLM graph attention for TCM prescription recommendation
Heterogeneous and homogeneous symptom-herb graphs plus an LLM-enhanced causal mechanism (published, not granted).
Multi-task joint optimization for TCM prescription generation
A pretrained LLM predicts herb sequences and herb sets jointly, with extra weight on rare herbs (published, not granted).
LLM-based intelligent acupuncture diagnosis agent
Acupuncture knowledge bases, graph, vector search and reranking, with an LLM agent that classifies questions and asks follow-ups (published, not granted).
TCM diagnosis-and-treatment platform combining an LLM with dual-channel retrieval
LLM plus TCM knowledge-graph dual-channel retrieval for inheriting master-physician experience and generating formulas (published, not granted).
Knowledge-graph and LLM system for TCM syndrome-differentiation QA
Chunks classics and cases, extracts a graph, then retrieves local and global keywords for syndrome-differentiation answers (published, not granted).
Structured evidence-subgraph RAG for TCM herb recommendation
Multi-hop evidence subgraphs plus Hopfield associative retrieval for herb recommendation, with graph constraints against hallucination (published, not granted).
Knowledge-enhanced generation of TCM case commentaries
Retrieves knowledge-graph subgraphs, then writes case notes along etiology, treatment principles, formulas, and prognosis (published, not granted).
Knowledge-distillation and RL for oncology TCM prescription recommendation
Distills GPT-4o into a Qwen student, then DPO, for interpretable oncology TCM prescriptions (published, not granted).
Multimodal knowledge-graph construction from TCM texts and records
An LLM extracts triples from TCM texts, then fuses a real-world case layer into a multimodal graph (published, not granted).
Knowledge-graph recommendation of classic TCM formulas
Maps colloquial symptoms to terms, filters by constitution, and recommends classic formulas (granted).
Knowledge-graph and case-enhanced RAG for intelligent TCM inquiry
Leiden subgraph search plus hybrid global/local recall, then a single LLM response (published, not granted).
AI-LLM method and system for intelligent TCM inquiry
Multimodal inquiry fusing face/body/tongue images, speech prosody, and a disease library (granted).
Multimodal knowledge-graph and LLM system for TCM rehabilitation diagnosis
Aligns tongue, pulse, and inquiry features with knowledge-graph embeddings before a rehabilitation expert LLM (published, not granted).
LLM-based Chinese-medicine prescription recommendation with soft prompts
Soft prompts and cross-attention feed patient text into an LLM for controllable herb recommendations (published, not granted).
LLM method for learning famous-physician TCM formulas
Aligns classical-Chinese case terms with vernacular text using models such as Qwen (published, not granted).
LLM and knowledge-graph method for TCM formula compatibility
Pairs a TCM LLM with a knowledge graph for formula compatibility (published, not granted).
LLM-based Chinese-materia-medica question answering
Chinese-materia-medica QA fine-tuned from Baichuan2-7B-Chat (granted February 2025).
Building a TCM QA system from a large language model and a knowledge graph
Closed loop of LLM generation and knowledge-graph completion for TCM QA (published, not granted).
AI-large-model system for intelligent TCM prescription recommendation
Trains an AI large model on medical records to recommend personalized TCM prescriptions, wired into a hospital IS (published, not granted).
Long-document retrieval-augmented generation for TCM question answering
Long-document RAG (expansion, recall, reranking, source citations) for TCM QA (granted July 2024).