How to Build a Smart Knowledge Graph for Specific Topics

Oracle's patent describes a method for automatically expanding a small, topic-specific knowledge graph by intelligently pulling relevant information from a much larger knowledge graph, helping applications understand user requests better.

What it covers

This patent describes a computer-implemented method for creating a specialized knowledge graph for a particular application. It starts by taking information from a user's own data, like conversations or documents, to build a basic "seed graph" of important entities and how they connect (Claim 1). The system then looks for groups of related entities in this seed graph, called "weakly connected components" (Claim 1). It maps these entities to a much larger, existing "reference knowledge graph." Using a "finite state machine," it explores the reference graph from these mapped points to find new, potentially relevant entities and connections (Claim 1). A special "priority function" then scores these potential additions, considering factors like how many weakly connected components are present and the overall "graph density" (Claim 1). The highest-scoring new entities and links are added to the seed graph, making it a more complete "customized knowledge graph." Finally, this expanded graph is used to understand what a user means when they ask a question or give a command, and then provide a suitable answer (Claim 1). For example, if a user dataset is about a specific company's products, the system could expand this with general product knowledge from a large reference graph to better answer customer service questions.

What it doesn't cover

• —Does not cover building a knowledge graph without first creating a "seed graph" from a specific "user dataset" (Claim 1).
• —Does not cover expanding a knowledge graph without using a separate, pre-existing "reference knowledge graph" for additional information (Claim 1).
• —Does not cover traversing the reference knowledge graph without using a "finite state machine" to guide the search for new entities (Claim 1).
• —Does not cover selecting new entities and links without a "priority function" that specifically considers the number of "weakly connected components" and the "graph density" (Claim 1).
• —Does not cover knowledge graph generation if it is not ultimately used for "determining a user intent" and "providing a response" to an utterance (Claim 1).

The clever bit

The clever part is how the system intelligently expands a small, focused knowledge graph by using a priority function. This function doesn't just add nearby information; it specifically evaluates new entities based on how they improve the overall structure and interconnectedness (measured by weakly connected components and graph density) of the growing graph, ensuring relevance and coherence.

Why it matters

This technology is important for making artificial intelligence systems smarter and more useful in specific areas. By customizing knowledge graphs, applications can better understand nuanced user requests and provide more accurate, context-aware responses. This is critical for improving the performance of chatbots, virtual assistants, and specialized search tools in fields like customer support, healthcare, or financial services.

Real-world examples

• 1.Customer service chatbots that answer specific product questions
• 2.Virtual assistants that understand industry-specific jargon
• 3.Enterprise search engines that provide highly relevant internal documents
• 4.Healthcare systems that interpret patient queries based on medical knowledge bases