Knowledge Graph Gap Filling via Expert Questioning
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Knowledge graphs face challenges in extracting and managing diverse, tacit knowledge due to their complexity and domain-specific nature, with existing methods being ineffective in identifying gaps and eliciting relevant information from experts, particularly in industrial settings where valuable knowledge is often not documented.
Innovation Solution
A computer-based method that identifies missing links in knowledge graphs by generating inquisitive and contextually relevant questions for experts, using node embeddings and natural language processing to capture semantic similarities and elicit responses, which are then evaluated and used to populate the graphs with informative answers.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Quantity of substance
If knowledge graphs are populated based on domain-specific ontology, then the knowledge graph can store diverse knowledge, but it becomes not very understandable for those estranged to that domain
Solution Approach 1:
The patent introduces a natural language processing intermediary layer that translates between domain-specific ontology and general understanding. The system uses NLP to process and interpret knowledge in a way that bridges the gap between specialized domain knowledge and general comprehension, allowing experts to contribute domain-specific knowledge while making it accessible to non-experts.
2Quantity of substance
If information is extracted from documents and databases, then knowledge graphs can be populated, but valuable tacit knowledge remains encapsulated in experts' minds and is not captured
Solution Approach 1:
The patent implements a feedback mechanism where the system identifies gaps in the knowledge graph and automatically generates questions to elicit tacit knowledge from experts. The experts' responses are fed back into the knowledge graph, continuously enriching it with previously undocumented tacit knowledge while maintaining a record of what has been learned.
Solution Approach 2:
The patent replaces manual knowledge extraction methods with an automated question-generation and answer-collection system. Instead of relying on experts to voluntarily document their knowledge or on manual interviews, the system automatically identifies knowledge gaps, formulates appropriate questions, and collects responses, thereby capturing tacit knowledge at scale.
3Extent of automation
If knowledge graph completion is attempted by treating knowledge graphs as collection of triples, then completion methods can be applied, but there is no robust mechanism to find out if the knowledge graph is incomplete
Solution Approach 1:
The system employs feedback loops where the completion process continuously evaluates the knowledge graph's state, identifies missing information through gap analysis, and adjusts the completion strategy accordingly. This feedback mechanism enables the system to detect incompleteness and systematically address it through targeted question generation and expert consultation.
4Ease of operation
If experts are asked to fill in relevant information using conversational agent, then the process is eased for experts not adept at editing databases, but there exists no way to pose questions in a natural language to obtain missing information
Solution Approach 1:
The patent replaces complex database editing interfaces with natural language processing capabilities. Instead of requiring experts to navigate complex database schemas and editing tools, the system uses NLP to understand and process natural language questions and answers, automatically mapping them to the appropriate knowledge graph structures. This substitution dramatically simplifies expert participation while managing the underlying complexity through automated language processing.
Data Source
AI summary
A system and method for managing knowledge for knowledge graphs is provided. The method including identifying missing links in the knowledge graph; generating inquisitive and contextually relevant questions around the identified missing links for an expert of a domain; in an event no missing links are identified, inquisitive and contextually relevant questions are generated based on a topic and a textual paragraph of topic of interest provided by the expert; receiving response to the questions from the expert via a user interface; generating additional informative questions based on the domain or the response received from the expert or a combination thereof; evaluating the additional informative questions based on a ranking metric derived from a combination of parameters; and populating the missing links in the knowledge graphs, displayed on the user interface, with one or more responses generated corresponding to the evaluated additional informative questions having a highest ranking metric.