Construction and Application of Financial Knowledge Graphs: AI Key Technologies, Building Practices, and Real‑World Use Cases

Financial institutions have accumulated large amounts of structured and unstructured data. This article introduces Daguan's financial knowledge graph construction and application, covering AI key technologies, domain knowledge‑graph building practices, and concrete financial use cases.

Artificial Intelligence Key Technologies

Perception technology: sensors, edge computing, computer vision, OCR, speech and language understanding, IoT, enabling machines to perceive the world.

Understanding technology: modeling and using knowledge, e.g., GPT‑3 large‑scale language models that capture human knowledge.

Planning & reasoning technology: neural networks, reinforcement learning, allowing machines to plan and reason like humans.

Action technology: RPA, robotic arms, autonomous driving, bipedal robots that execute actions and receive feedback.

Knowledge graphs integrate these technologies, forming a brain‑like system that stores and applies knowledge to specific scenarios, turning raw data into actionable insights.

Daguan Intelligent Products

Document Intelligent Processing Platform – one‑stop OCR, key‑information extraction, and intelligent review for various document types.

Daguan Intelligent RPA – non‑intrusive integration with business systems, using perception to extract information and action to automate processes.

Yuanhai Knowledge‑Graph Platform – a unified repository for domain knowledge and expert experience.

Yuanhai Knowledge‑Graph Platform Features

Data management: supports structured and unstructured data, MySQL, PostgreSQL, DM, CSV, Excel, etc.

Multi‑graph management: create, modify, delete graphs; per‑graph data source, function, algorithm, model, and permission management.

Graph construction & editing: mapping‑based, triple‑based, extraction‑based construction; CRUD for entities and relationships.

Graph algorithms: Spark, GraphX, deep‑learning‑based algorithms.

Fine‑grained RBAC permission control.

Knowledge extraction: end‑to‑end pipeline for entity, relation, and event extraction.

Knowledge‑Graph Construction Techniques

1. Schema Design

Define entity types, attributes, inheritance, and directed relationships; design iteratively with domain experts to match business needs.

2. Knowledge Extraction

Extract information from structured and unstructured data.

Mapping‑based construction : directly map structured sources (e.g., stock data, financial statements) into the graph.

Extraction‑based construction : apply NLP techniques to documents, news, emails, etc., to extract entities and relations.

Entity extraction (NER) methods include rule‑based, machine‑learning, deep‑learning, and hybrid approaches such as BiLSTM‑CRF, BERT‑based fine‑tuning, and CNN‑based remote supervision.

Relation extraction methods include rule‑based patterns, supervised classifiers, bootstrapping, and distant supervision using existing triples.

3. Knowledge Fusion

Merge structures from heterogeneous sources, ensuring semantic and business consistency.

4. Knowledge Storage

Use graph databases (Neo4j, JanusGraph) for small scale; for large scale, employ distributed storage (HBase, JanusGraph) combined with Elasticsearch, Spark, Kafka, and optionally store raw documents in HDFS.

5. Knowledge Application

Intelligent tagging of financial news using graph‑linked concepts.

Recommendation and matchmaking based on knowledge‑graph similarity.

Policy graphs to interpret governmental financial regulations.

Executive‑investment relationship mining for corporate governance analysis.

Semantic search, reasoning, Q&A, and real‑time reporting powered by the graph.

Financial Knowledge‑Graph Application Practices

Intelligent labeling of financial news for multi‑dimensional tagging and recommendation.

Knowledge‑graph‑driven recommendation and matchmaking for intellectual property data.

Policy graphs to help institutions understand financial regulations.

Mining potential investment relationships of senior executives.

Knowledge‑Graph Forum Registration (Promotional)

On December 19, 9:00‑12:00, a knowledge‑graph forum hosted by Alibaba senior algorithm expert Zhang Wei will feature speakers from Baidu, Alibaba, Meituan, and Beike. Scan the QR code to register.

For more AI‑related articles and community links, see the recommended reading list at the end of the original post.