Ontology Construction via AI-Mediated Crowdsourcing
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Modeling high-level activities in intelligent traffic management systems is challenging due to the need for diverse and contextual descriptions of events like traffic jams and accidents, which vary by culture and perspective, requiring collective intelligence and efficient information representation.
Innovation Solution
The system employs AI-mediated crowdsourcing and concept mining to collect and summarize natural-language inputs from crowdsourcing workers, using natural language processing to generate triples for augmenting knowledge graphs, integrating human input with machine learning for effective ontology construction.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Speed
If traditional automated methods are used for ontology construction, then processing speed is improved, but understanding accuracy of high-level activities deteriorates
Solution Approach 1:
The patent merges automated event detection systems with crowdsourcing platforms to create a hybrid ontology construction system. Automated methods quickly identify and categorize events in traffic scenes, while crowdsourced workers provide contextual annotations and semantic descriptions. This combination resolves the contradiction by maintaining high processing speed through automation while achieving high understanding accuracy through human contextual input.
Solution Approach 2:
The patent introduces an intermediary layer that bridges automated event detection and final ontology construction. This intermediary process involves crowdsourcing workers who act as mediators, translating machine-detected events into human-understandable semantic annotations. The intermediary layer preserves the speed benefits of automation while incorporating the accuracy benefits of human understanding.
2Adaptability or versatility
If diverse cultural perspectives are incorporated into event modeling, then ontology completeness is improved, but system complexity increases
Solution Approach 1:
The patent segments the ontology construction process into distinct modules: automated event detection, crowdsourcing annotation collection, and knowledge graph integration. Each module handles specific aspects of diverse cultural perspectives independently. This segmentation allows the system to incorporate multiple cultural viewpoints through structured crowdsourcing tasks without creating unmanageable overall system complexity.
Solution Approach 2:
The patent creates a universal crowdsourcing framework that can handle diverse cultural perspectives through standardized annotation schemas. The same platform and process structure accommodates different cultural contexts, event types, and description styles, achieving ontology completeness across cultures while maintaining consistent system architecture that prevents complexity escalation.
3Measurement precision
If manual annotation by experts is used, then annotation quality is improved, but time consumption increases
Solution Approach 1:
The patent implements a self-service crowdsourcing model where workers autonomously annotate events based on their own knowledge and cultural context, without requiring continuous expert supervision or manual review of each annotation. This self-service approach maintains high annotation quality through worker expertise while dramatically reducing time consumption compared to traditional manual expert annotation processes.
Solution Approach 2:
The patent incorporates feedback mechanisms where crowd workers receive guidance and validation during the annotation process, and where high-quality annotations from the community are reinforced. This feedback loop maintains annotation quality comparable to expert-level work while distributing the time investment across many workers, reducing overall time consumption compared to centralized expert annotation.
Data Source
AI summary
Methods and system of building and augmenting a knowledge graph regarding ontology of events occurring in images. Image data corresponding to a plurality of scenes captured by one or more cameras is received. A knowledge graph is built with event-based ontology data corresponding to events occurring in the plurality of scenes. One or more of the scenes is displayed to a plurality of crowdsourcing workers which provide natural-language input including event-based semantic annotations corresponding to the scene. Using natural language processing on the input, triples are generated. The knowledge graph is augmented with the generated triples to yield an augmented knowledge graph for use in determining event-based ontology associated with the plurality of scenes.