Knowledge Graph and RAG Applications in 360 Document Cloud: Challenges and Solutions

The presentation begins by outlining the business pain points of rapidly growing unstructured data, including storage, access, security, and low utilization, as well as challenges faced when applying large language models such as lack of domain expertise, data leakage risks, and limited contextual understanding.

It then describes the advantages of 360 Document Cloud, highlighting its ability to store massive high‑quality enterprise data, enforce multi‑level permission controls, and capture user behavior for context‑aware services.

The intelligent document solution is introduced in three layers: (1) deep document comprehension and summarization, (2) fast knowledge retrieval from massive corpora, and (3) accurate answer generation, enabling AI assistants, smart recommendations, and various downstream tasks such as translation and summarization.

Subsequently, the role of Knowledge Graphs (KG) in Retrieval‑Augmented Generation (RAG) is examined. KG provides unified management of heterogeneous data, supports semantic organization, entity linking, and structured storage of complex document hierarchies, thereby improving recall, intent recognition, prompt assembly, and result verification.

Practical KG‑enhanced workflows are detailed, including chunking and embedding strategies (e.g., M3E, Text2Vec, E5), model selection (360 Zhihui, ChatGLM, Llama2, ChatGPT), and evaluation of combined pipelines. The discussion also covers KG‑driven document standardization, fine‑tuning QA pair generation, and multi‑stage knowledge‑base management.

Finally, challenges and future directions are explored, such as controlled query rewriting with KG, improving KG construction automation, maintaining real‑time accuracy, schema generation at scale, and integrating KG as an independent retrieval source alongside LLMs.

The session concludes with a Q&A segment addressing query expansion evaluation and invites further exploration of AI‑enhanced document management.