Unsupervised Relation Detection Model Training via Web Mining
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Natural language understanding systems face challenges in scaling to new domains and languages due to the time-consuming and expensive process of manual annotation, and existing methods struggle with semantic inconsistency between systems and knowledge sources, limiting interoperability.
Innovation Solution
A relation detection model training solution that mines freely available resources from the World Wide Web to train a relationship detection model, using a knowledge graph to define semantic ontology and mine training examples, which scales to other domains and languages, and relies on the semantic space dictated by the domain knowledge graph for informational user requests.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If manual annotation is used for training data preparation, then model accuracy can be improved, but the process becomes time-consuming and expensive
Solution Approach 1:
The patent uses web search snippets as copies of natural language data that already contain entity-relation information. Instead of manually creating training examples, the system copies relevant information from existing web content, automatically processing it into training data that maintains accuracy while eliminating manual annotation time and cost
Solution Approach 2:
The system performs self-service by automatically mining training data from web search results without human intervention. The unsupervised learning approach allows the model to autonomously identify and extract relation patterns from snippets, eliminating the need for manual annotation while maintaining effective training
2Reliability
If semantic template-based systems are used, then intent determination can be achieved, but the systems do not scale well to other domains and languages
Solution Approach 1:
The patent creates a universal relation detection model that can handle multiple domains and languages simultaneously. By training on diverse web data from various domains and languages, the model becomes multi-functional, able to detect relations across different contexts without requiring domain-specific semantic templates, thus achieving both reliability and scalability
3Measurement precision
If system-specific semantic space is aligned with knowledge graph, then slot filling performance improves, but semantic inconsistency with other systems occurs
Solution Approach 1:
Instead of aligning the system's semantic space with the knowledge graph, the patent inverts the approach by having the system learn directly from the knowledge graph's structure and data. The model trains on entity-relation patterns from the knowledge graph, allowing it to naturally understand and query the knowledge graph without semantic space alignment issues, thereby improving both slot filling performance and interoperability
Data Source
AI summary
A relation detection model training solution. The relation detection model training solution mines freely available resources from the World Wide Web to train a relationship detection model for use during linguistic processing. The relation detection model training system searches the web for pairs of entities extracted from a knowledge graph that are connected by a specific relation. Performance is enhanced by clipping search snippets to extract patterns that connect the two entities in a dependency tree and refining the annotations of the relations according to other related entities in the knowledge graph. The relation detection model training solution scales to other domains and languages, pushing the burden from natural language semantic parsing to knowledge base population. The relation detection model training solution exhibits performance comparable to supervised solutions, which require design, collection, and manual labeling of natural language data.