Wikipedia Question Answering via Multi-Document Segmentation

The patent segments Wikipedia content into multiple document types (fulltext documents, section title documents, info-box documents, category documents, definition statement documents) rather than converting everything to a single standardized knowledge base format. This segmentation allows preserving the original information structure while enabling targeted access to different types of information, thus reducing information loss while still providing ambiguity reduction through structured organization.

The patent introduces a new dimension by creating multiple parallel document representations instead of a single knowledge base conversion path. Each document type serves a different function and preserves different aspects of the original Wikipedia content, allowing the system to access information from multiple dimensions simultaneously, thereby avoiding the information loss inherent in single-path conversion.

The patent segments the question processing into multiple parallel paths, each handling different document types (fulltext, section titles, info-boxes, categories, definition statements). Each path extracts answers independently based on the characteristics of that document type, avoiding the need to convert the entire natural language question into a single standardized knowledge base expression, thus reducing conversion ambiguity while maintaining precision.

The patent applies partial action by extracting answers from only the relevant document types for each specific question, rather than converting and searching all Wikipedia content. For example, definition statement documents are used for definition-type questions, while info-box documents are used for factual attribute questions, reducing the ambiguity of full conversion while maintaining precise answer extraction for each question type.

The patent merges multiple document types with different structures into a unified question answering system. It combines fulltext documents (unstructured), section title documents (semi-structured), info-box documents (structured), category documents (hierarchical), and definition statement documents (structured) into a single system that processes questions through multiple parallel paths, thereby preserving both the precision of structured information and the completeness of unstructured information.

The patent creates a universal question answering apparatus that handles multiple document types and question types through a unified framework. The system can process different kinds of questions (definition questions, factual questions, descriptive questions) using appropriate document types, making the system multi-functional while avoiding information loss from excluding any document type.

The patent creates simplified copies of Wikipedia content in multiple document types rather than performing a complete and complex conversion to a single standardized knowledge base. Each document type is a targeted copy that preserves specific aspects of the original content needed for different question types, reducing the complexity of knowledge base construction while maintaining standardization where needed.

The patent segments the knowledge representation into multiple specialized document types rather than creating one comprehensive standardized knowledge base. This segmentation reduces the complexity of building and maintaining the knowledge representation system, as each document type can be processed and maintained independently with simpler rules, while collectively providing comprehensive coverage.